[management, proxy] Management-owned LLM pricing: file-backed defaults + (#6965)

2026-08-30 19:41:30 +02:00 · 2026-07-31 20:52:56 +02:00
parent feecb993f4
commit 0780a806f2
47 changed files with 3196 additions and 1540 deletions
--- a/proxy/internal/llm/bedrock_model.go
+++ b/proxy/internal/llm/bedrock_model.go
@@ -1,38 +0,0 @@
-package llm
-
-import (
-	"regexp"
-	"strings"
-)
-
-// bedrockRegionPrefixes are the cross-region inference-profile prefixes that
-// front a Bedrock model id (e.g. "eu.anthropic.claude-...").
-var bedrockRegionPrefixes = []string{"us.", "eu.", "apac.", "global."}
-
-// bedrockVersionSuffix matches the trailing "-vN[:N]" or "-YYYYMMDD-vN[:N]"
-// version/throughput suffix of a Bedrock model id.
-var bedrockVersionSuffix = regexp.MustCompile(`-(\d{8}-)?v\d+(:\d+)?$`)
-
-// NormalizeBedrockModel strips an ARN wrapper, a cross-region inference-profile
-// prefix, and the version/throughput suffix from a Bedrock model id so it
-// matches the catalog/pricing key, e.g.
-// "eu.anthropic.claude-sonnet-4-5-20250929-v1:0" -> "anthropic.claude-sonnet-4-5"
-// and the inference-profile ARN's last segment likewise. It is the single
-// source of truth shared by the request parser (which normalizes the request
-// model from the URL path) and the router (which normalizes the operator's
-// registered Bedrock model ids so both sides compare equal).
-func NormalizeBedrockModel(modelID string) string {
-	m := modelID
-	if strings.HasPrefix(m, "arn:") {
-		if i := strings.LastIndex(m, "/"); i >= 0 {
-			m = m[i+1:]
-		}
-	}
-	for _, p := range bedrockRegionPrefixes {
-		if strings.HasPrefix(m, p) {
-			m = m[len(p):]
-			break
-		}
-	}
-	return bedrockVersionSuffix.ReplaceAllString(m, "")
-}
--- a/proxy/internal/llm/bedrock_model_test.go
+++ b/proxy/internal/llm/bedrock_model_test.go
@@ -1,23 +0,0 @@
-package llm
-
-import (
-	"testing"
-
-	"github.com/stretchr/testify/require"
-)
-
-func TestNormalizeBedrockModel(t *testing.T) {
-	cases := map[string]string{
-		"eu.anthropic.claude-sonnet-4-5-20250929-v1:0": "anthropic.claude-sonnet-4-5",
-		"us.anthropic.claude-haiku-4-5":                "anthropic.claude-haiku-4-5",
-		"us.anthropic.claude-opus-4-8-20250101-v1:0":   "anthropic.claude-opus-4-8",
-		"anthropic.claude-sonnet-4-5-20250929-v1:0":    "anthropic.claude-sonnet-4-5",
-		"meta.llama3-3-70b-instruct-v1:0":              "meta.llama3-3-70b-instruct",
-		"amazon.nova-pro-v1:0":                         "amazon.nova-pro",
-		// Inference-profile ARN — model id lives in the last path segment.
-		"arn:aws:bedrock:eu-central-1:123456789012:inference-profile/eu.anthropic.claude-sonnet-4-5-20250929-v1:0": "anthropic.claude-sonnet-4-5",
-	}
-	for in, want := range cases {
-		require.Equal(t, want, NormalizeBedrockModel(in), "normalize %q", in)
-	}
-}
--- a/proxy/internal/llm/fixtures/pricing.yaml
+++ b/proxy/internal/llm/fixtures/pricing.yaml
@@ -1,59 +0,0 @@
-# Realistic-pricing starter for llm_observability. Drop this into the
-# directory you point the proxy at via --plugin-data-dir, then reference it
-# from the target's plugin config:
-#
-#   plugins:
-#     - id: llm_observability
-#       enabled: true
-#       params:
-#         pricing_path: pricing.yaml
-#
-# Values are USD per 1_000 tokens. Public list prices drift; treat this as a
-# starting point and keep your production copy current.
-
-openai:
-  # GPT-5 family
-  gpt-5:
-    input_per_1k: 0.00125
-    output_per_1k: 0.01
-  gpt-5-mini:
-    input_per_1k: 0.00025
-    output_per_1k: 0.002
-  gpt-5-nano:
-    input_per_1k: 0.00005
-    output_per_1k: 0.0004
-  gpt-5.4:
-    input_per_1k: 0.00125
-    output_per_1k: 0.01
-  # GPT-4o family
-  gpt-4o:
-    input_per_1k: 0.0025
-    output_per_1k: 0.01
-  gpt-4o-mini:
-    input_per_1k: 0.00015
-    output_per_1k: 0.0006
-  # Embeddings
-  text-embedding-3-large:
-    input_per_1k: 0.00013
-    output_per_1k: 0
-  text-embedding-3-small:
-    input_per_1k: 0.00002
-    output_per_1k: 0
-
-anthropic:
-  # Claude 4.x family
-  claude-opus-4-7:
-    input_per_1k: 0.015
-    output_per_1k: 0.075
-  claude-sonnet-4-7:
-    input_per_1k: 0.003
-    output_per_1k: 0.015
-  claude-sonnet-4-6:
-    input_per_1k: 0.003
-    output_per_1k: 0.015
-  claude-sonnet-4-5:
-    input_per_1k: 0.003
-    output_per_1k: 0.015
-  claude-haiku-4-5:
-    input_per_1k: 0.0008
-    output_per_1k: 0.004
--- a/proxy/internal/llm/model.go
+++ b/proxy/internal/llm/model.go
@@ -0,0 +1,21 @@
+package llm
+
+import (
+	sharedllm "github.com/netbirdio/netbird/shared/llm"
+)
+
+// NormalizeBedrockModel strips an ARN wrapper, a cross-region inference-profile
+// prefix, and the version/throughput suffix from a Bedrock model id so it
+// matches the catalog/pricing key. Thin delegate to the shared implementation
+// (shared/llm), which management also uses at synthesis time so both sides of
+// the pricing / routing contract normalize identically.
+func NormalizeBedrockModel(modelID string) string {
+	return sharedllm.NormalizeBedrockModel(modelID)
+}
+
+// NormalizeVertexModel strips the "@version" suffix from a Vertex AI model id
+// so it matches the catalog/pricing key. Thin delegate to shared/llm, kept
+// beside NormalizeBedrockModel for the same contract reason.
+func NormalizeVertexModel(modelID string) string {
+	return sharedllm.NormalizeVertexModel(modelID)
+}
--- a/proxy/internal/llm/pricing/defaults_coverage_test.go
+++ b/proxy/internal/llm/pricing/defaults_coverage_test.go
@@ -1,65 +0,0 @@
-package pricing
-
-import (
-	"testing"
-
-	"github.com/stretchr/testify/assert"
-	"github.com/stretchr/testify/require"
-)
-
-// TestDefaultTable_FirstPartyModelCoverage guards the embedded defaults against
-// silent drift/gaps: every metered first-party model the management catalog
-// enumerates must resolve to a price, and a few rates that previously drifted
-// are pinned to their LiteLLM-validated values. Keep this list in step with the
-// catalog (management/server/agentnetwork/catalog) when adding models.
-func TestDefaultTable_FirstPartyModelCoverage(t *testing.T) {
-	tbl := DefaultTable()
-	require.NotNil(t, tbl, "embedded default pricing table must load")
-
-	mustPrice := map[string][]string{
-		// openai parser covers openai_api, azure_openai_api, and mistral_api.
-		"openai": {
-			"gpt-5.5", "gpt-5.5-pro", "gpt-5.4", "gpt-5.4-mini", "gpt-5.4-nano",
-			"gpt-5.3-codex", "gpt-5.3-chat-latest", "o4-mini",
-			"gpt-4.1", "gpt-4.1-mini", "gpt-4.1-nano", "gpt-4o", "gpt-4o-mini",
-			"gpt-4-turbo", "gpt-3.5-turbo", "gpt-35-turbo",
-			"text-embedding-3-large", "text-embedding-3-small",
-			"mistral-large-latest", "mistral-medium-3-5", "codestral-2508",
-			"ministral-8b-latest", "mistral-embed",
-		},
-		"anthropic": {
-			"claude-fable-5", "claude-opus-5", "claude-opus-4-8", "claude-opus-4-7", "claude-opus-4-6",
-			"claude-opus-4-1", "claude-sonnet-4-6", "claude-sonnet-4-5", "claude-haiku-4-5",
-		},
-		// bedrock keys are the normalized ids the request parser emits.
-		"bedrock": {
-			"anthropic.claude-opus-5", "anthropic.claude-opus-4-8", "anthropic.claude-opus-4-7", "anthropic.claude-opus-4-6",
-			"anthropic.claude-opus-4-1", "anthropic.claude-sonnet-4-6", "anthropic.claude-sonnet-4-5",
-			"anthropic.claude-haiku-4-5", "meta.llama3-3-70b-instruct",
-			"amazon.nova-pro", "amazon.nova-lite", "amazon.nova-micro", "amazon.nova-2-lite",
-		},
-	}
-	for provider, models := range mustPrice {
-		for _, m := range models {
-			_, ok := tbl.Cost(provider, m, 1000, 1000, 0, 0)
-			assert.True(t, ok, "%s/%s must be priced in the embedded defaults", provider, m)
-		}
-	}
-
-	// Pin per-direction rates independently (input-only then output-only) so a
-	// swap or skew of input<->output that preserves the combined total is still
-	// caught — these are rates that previously drifted or are easy to mis-enter.
-	in, ok := tbl.Cost("openai", "gpt-5.4", 1000, 0, 0, 0)
-	require.True(t, ok)
-	assert.InDelta(t, 0.0025, in, 1e-9, "gpt-5.4 input = 0.0025 per 1k")
-	out, ok := tbl.Cost("openai", "gpt-5.4", 0, 1000, 0, 0)
-	require.True(t, ok)
-	assert.InDelta(t, 0.015, out, 1e-9, "gpt-5.4 output = 0.015 per 1k")
-
-	in, ok = tbl.Cost("bedrock", "anthropic.claude-sonnet-4-5", 1000, 0, 0, 0)
-	require.True(t, ok)
-	assert.InDelta(t, 0.003, in, 1e-9, "bedrock sonnet-4-5 input = 0.003 per 1k")
-	out, ok = tbl.Cost("bedrock", "anthropic.claude-sonnet-4-5", 0, 1000, 0, 0)
-	require.True(t, ok)
-	assert.InDelta(t, 0.015, out, 1e-9, "bedrock sonnet-4-5 output = 0.015 per 1k")
-}
--- a/proxy/internal/llm/pricing/defaults_pricing.yaml
+++ b/proxy/internal/llm/pricing/defaults_pricing.yaml
@@ -1,298 +0,0 @@
-# Embedded default pricing for llm_observability. Compiled into the proxy
-# binary via go:embed in pricing.go; cost annotation works out of the box
-# without any operator action.
-#
-# Operators override entries by dropping a pricing.yaml into --plugin-data-dir
-# (or whichever basename is given via params.pricing_path). The override file
-# only needs entries the operator wants to change; missing entries fall
-# through to these defaults.
-#
-# Values are USD per 1_000 tokens. Public list prices drift; ship a fresh
-# binary or override individual entries via the override file as needed.
-#
-# Optional cache fields:
-#   cached_input_per_1k    OpenAI: rate for prompt_tokens_details.cached_tokens
-#                          (a SUBSET of prompt_tokens). Typically 0.5x input.
-#                          Absent → cached portion bills at input_per_1k.
-#   cache_read_per_1k      Anthropic: rate for cache_read_input_tokens
-#                          (ADDITIVE to input_tokens). Typically 0.1x input.
-#                          Absent → cache reads bill at input_per_1k.
-#   cache_creation_per_1k  Anthropic: rate for cache_creation_input_tokens
-#                          (ADDITIVE to input_tokens). Typically 1.25x input.
-#                          Absent → cache writes bill at input_per_1k.
-
-openai:
-  # OpenAI + OpenAI-compatible providers (openai_api, azure_openai_api,
-  # mistral_api, and the openai-parser gateways) all emit llm.provider="openai",
-  # so their models are priced here. Kept in sync with the management catalog;
-  # rates cross-checked against LiteLLM model_prices_and_context_window.json.
-
-  # GPT-5.x family — cache reads 10% of input (0.1x).
-  gpt-5.5:
-    input_per_1k: 0.005
-    output_per_1k: 0.03
-    cached_input_per_1k: 0.0005
-  gpt-5.5-pro:
-    input_per_1k: 0.03
-    output_per_1k: 0.18
-    cached_input_per_1k: 0.003
-  gpt-5.4:
-    input_per_1k: 0.0025
-    output_per_1k: 0.015
-    cached_input_per_1k: 0.00025
-  gpt-5.4-pro:
-    input_per_1k: 0.03
-    output_per_1k: 0.18
-    cached_input_per_1k: 0.003
-  gpt-5.4-mini:
-    input_per_1k: 0.00075
-    output_per_1k: 0.0045
-    cached_input_per_1k: 0.000075
-  gpt-5.4-nano:
-    input_per_1k: 0.0002
-    output_per_1k: 0.00125
-    cached_input_per_1k: 0.00002
-  gpt-5.3-codex:
-    input_per_1k: 0.00175
-    output_per_1k: 0.014
-    cached_input_per_1k: 0.000175
-  gpt-5.3-chat-latest:
-    input_per_1k: 0.00175
-    output_per_1k: 0.014
-    cached_input_per_1k: 0.000175
-  # GPT-5 (2025) family — kept for gateway requests using the unsuffixed ids.
-  gpt-5:
-    input_per_1k: 0.00125
-    output_per_1k: 0.01
-    cached_input_per_1k: 0.000125
-  gpt-5-mini:
-    input_per_1k: 0.00025
-    output_per_1k: 0.002
-    cached_input_per_1k: 0.000025
-  gpt-5-nano:
-    input_per_1k: 0.00005
-    output_per_1k: 0.0004
-    cached_input_per_1k: 0.000005
-  o4-mini:
-    input_per_1k: 0.0011
-    output_per_1k: 0.0044
-    cached_input_per_1k: 0.000275
-  # GPT-4.1 family — cache reads 25% of input.
-  gpt-4.1:
-    input_per_1k: 0.002
-    output_per_1k: 0.008
-    cached_input_per_1k: 0.0005
-  gpt-4.1-mini:
-    input_per_1k: 0.0004
-    output_per_1k: 0.0016
-    cached_input_per_1k: 0.0001
-  gpt-4.1-nano:
-    input_per_1k: 0.0001
-    output_per_1k: 0.0004
-    cached_input_per_1k: 0.000025
-  # GPT-4o family — cache reads 50% of input (0.5x).
-  gpt-4o:
-    input_per_1k: 0.0025
-    output_per_1k: 0.01
-    cached_input_per_1k: 0.00125
-  gpt-4o-mini:
-    input_per_1k: 0.00015
-    output_per_1k: 0.0006
-    cached_input_per_1k: 0.000075
-  # Older GPT — no prompt caching.
-  gpt-4-turbo:
-    input_per_1k: 0.01
-    output_per_1k: 0.03
-  gpt-3.5-turbo:
-    input_per_1k: 0.0005
-    output_per_1k: 0.0015
-  gpt-35-turbo: # Azure deployment alias of gpt-3.5-turbo
-    input_per_1k: 0.0005
-    output_per_1k: 0.0015
-  # Embeddings — no caching, no output tokens.
-  text-embedding-3-large:
-    input_per_1k: 0.00013
-    output_per_1k: 0
-  text-embedding-3-small:
-    input_per_1k: 0.00002
-    output_per_1k: 0
-
-  # Mistral (mistral_api) — routed via the openai parser; no prompt caching.
-  mistral-large-latest:
-    input_per_1k: 0.0005
-    output_per_1k: 0.0015
-  mistral-medium-latest:
-    input_per_1k: 0.0004
-    output_per_1k: 0.002
-  mistral-medium-3-5:
-    input_per_1k: 0.0015
-    output_per_1k: 0.0075
-  mistral-small-latest:
-    input_per_1k: 0.00006
-    output_per_1k: 0.00018
-  magistral-medium-latest:
-    input_per_1k: 0.002
-    output_per_1k: 0.005
-  magistral-small-latest:
-    input_per_1k: 0.0005
-    output_per_1k: 0.0015
-  devstral-medium-latest:
-    input_per_1k: 0.0004
-    output_per_1k: 0.002
-  devstral-small-latest:
-    input_per_1k: 0.0001
-    output_per_1k: 0.0003
-  codestral-2508:
-    input_per_1k: 0.0003
-    output_per_1k: 0.0009
-  codestral-latest:
-    input_per_1k: 0.001
-    output_per_1k: 0.003
-  ministral-3-14b-2512:
-    input_per_1k: 0.0002
-    output_per_1k: 0.0002
-  ministral-8b-latest:
-    input_per_1k: 0.00015
-    output_per_1k: 0.00015
-  ministral-3-3b-2512:
-    input_per_1k: 0.0001
-    output_per_1k: 0.0001
-  mistral-embed:
-    input_per_1k: 0.0001
-    output_per_1k: 0
-
-  # Kimi / Moonshot AI (kimi_api) — OpenAI-compatible /v1 endpoint. Moonshot
-  # reports cache hits OpenAI-style when present; cached input is 10% of
-  # input ($0.30 vs $3.00 per MTok). kimi-k3 is the only model the platform
-  # serves newer accounts (K2-era ids and kimi-latest 404), matching the
-  # management catalog.
-  kimi-k3:
-    input_per_1k: 0.003
-    output_per_1k: 0.015
-    cached_input_per_1k: 0.0003
-
-anthropic:
-  # Claude 4.x family — cache reads ≈10% of input, cache writes ≈125% of input.
-  # Pricing source: Anthropic's current published rates per million tokens,
-  # divided by 1000 for the per-1k figures stored here.
-  claude-fable-5:
-    input_per_1k: 0.010
-    output_per_1k: 0.050
-    cache_read_per_1k: 0.001
-    cache_creation_per_1k: 0.0125
-  claude-opus-5:
-    input_per_1k: 0.005
-    output_per_1k: 0.025
-    cache_read_per_1k: 0.0005
-    cache_creation_per_1k: 0.00625
-  claude-opus-4-8:
-    input_per_1k: 0.005
-    output_per_1k: 0.025
-    cache_read_per_1k: 0.0005
-    cache_creation_per_1k: 0.00625
-  claude-opus-4-7:
-    input_per_1k: 0.005
-    output_per_1k: 0.025
-    cache_read_per_1k: 0.0005
-    cache_creation_per_1k: 0.00625
-  claude-opus-4-6:
-    input_per_1k: 0.005
-    output_per_1k: 0.025
-    cache_read_per_1k: 0.0005
-    cache_creation_per_1k: 0.00625
-  claude-opus-4-1:
-    input_per_1k: 0.015
-    output_per_1k: 0.075
-    cache_read_per_1k: 0.0015
-    cache_creation_per_1k: 0.01875
-  claude-sonnet-4-6:
-    input_per_1k: 0.003
-    output_per_1k: 0.015
-    cache_read_per_1k: 0.0003
-    cache_creation_per_1k: 0.00375
-  claude-sonnet-4-5:
-    input_per_1k: 0.003
-    output_per_1k: 0.015
-    cache_read_per_1k: 0.0003
-    cache_creation_per_1k: 0.00375
-  claude-haiku-4-5:
-    input_per_1k: 0.001
-    output_per_1k: 0.005
-    cache_read_per_1k: 0.0001
-    cache_creation_per_1k: 0.00125
-
-  # Kimi / Moonshot AI (kimi_api) via the Anthropic-compatible endpoint
-  # (/anthropic/v1/messages — the official Claude Code setup). Same rates
-  # as the OpenAI-shape entry above. "kimi-k3[1m]" is the model id some
-  # Claude Code guides set for the 1M-context alias; priced identically so
-  # cost metering doesn't silently skip those requests.
-  kimi-k3:
-    input_per_1k: 0.003
-    output_per_1k: 0.015
-    cache_read_per_1k: 0.0003
-  "kimi-k3[1m]":
-    input_per_1k: 0.003
-    output_per_1k: 0.015
-    cache_read_per_1k: 0.0003
-
-bedrock:
-  # AWS Bedrock model ids, normalised by the request parser (cross-region
-  # inference-profile prefix + version/throughput suffix stripped), e.g.
-  # eu.anthropic.claude-sonnet-4-5-20250929-v1:0 -> anthropic.claude-sonnet-4-5.
-  # Anthropic-on-Bedrock keeps the additive cache buckets (read ≈0.1x input,
-  # write ≈1.25x input); Nova / Llama report no cache, so cost is input+output.
-  anthropic.claude-opus-5:
-    input_per_1k: 0.005
-    output_per_1k: 0.025
-    cache_read_per_1k: 0.0005
-    cache_creation_per_1k: 0.00625
-  anthropic.claude-opus-4-8:
-    input_per_1k: 0.005
-    output_per_1k: 0.025
-    cache_read_per_1k: 0.0005
-    cache_creation_per_1k: 0.00625
-  anthropic.claude-opus-4-7:
-    input_per_1k: 0.005
-    output_per_1k: 0.025
-    cache_read_per_1k: 0.0005
-    cache_creation_per_1k: 0.00625
-  anthropic.claude-opus-4-6:
-    input_per_1k: 0.005
-    output_per_1k: 0.025
-    cache_read_per_1k: 0.0005
-    cache_creation_per_1k: 0.00625
-  anthropic.claude-opus-4-1:
-    input_per_1k: 0.015
-    output_per_1k: 0.075
-    cache_read_per_1k: 0.0015
-    cache_creation_per_1k: 0.01875
-  anthropic.claude-sonnet-4-6:
-    input_per_1k: 0.003
-    output_per_1k: 0.015
-    cache_read_per_1k: 0.0003
-    cache_creation_per_1k: 0.00375
-  anthropic.claude-sonnet-4-5:
-    input_per_1k: 0.003
-    output_per_1k: 0.015
-    cache_read_per_1k: 0.0003
-    cache_creation_per_1k: 0.00375
-  anthropic.claude-haiku-4-5:
-    input_per_1k: 0.001
-    output_per_1k: 0.005
-    cache_read_per_1k: 0.0001
-    cache_creation_per_1k: 0.00125
-  meta.llama3-3-70b-instruct:
-    input_per_1k: 0.00072
-    output_per_1k: 0.00072
-  amazon.nova-2-lite:
-    input_per_1k: 0.0003
-    output_per_1k: 0.0025
-  amazon.nova-pro:
-    input_per_1k: 0.0008
-    output_per_1k: 0.0032
-  amazon.nova-lite:
-    input_per_1k: 0.00006
-    output_per_1k: 0.00024
-  amazon.nova-micro:
-    input_per_1k: 0.000035
-    output_per_1k: 0.00014
--- a/proxy/internal/llm/pricing/pricing.go
+++ b/proxy/internal/llm/pricing/pricing.go
@@ -1,102 +1,30 @@
-// Package pricing implements the embedded-default + override pricing table
-// shared by middleware that converts LLM token usage into a USD cost
-// estimate. The table is hot-reloadable from a basename under the proxy
-// data directory; missing override files keep the embedded defaults so
-// cost annotation works without operator action.
+// Package pricing implements the pricing table and cost formula the
+// cost_meter middleware uses to convert LLM token usage into a USD cost
+// estimate. The table's content arrives from the management server inside
+// cost_meter's middleware config (synthesized from the catalog plus the
+// operator's stored per-provider prices) — the proxy carries no embedded
+// price list. Price updates ride the ordinary mapping push: a chain
+// rebuild constructs a fresh table, so there is nothing to reload.
 package pricing

 import (
-	"bytes"
-	"context"
-	_ "embed"
-	"errors"
 	"fmt"
-	"io"
-	"io/fs"
 	"math"
-	"path/filepath"
-	"regexp"
-	"strings"
-	"sync"
-	"sync/atomic"
-	"time"
-
-	log "github.com/sirupsen/logrus"
-	"go.opentelemetry.io/otel/attribute"
-	"go.opentelemetry.io/otel/metric"
-	"gopkg.in/yaml.v3"
 )

-//go:embed defaults_pricing.yaml
-var defaultPricingYAML []byte
-
-var (
-	defaultTableOnce sync.Once
-	defaultTablePtr  *Table
-)
-
-// DefaultTable returns the pricing table embedded in the binary. The result
-// is parsed once and shared; callers must not mutate the returned value.
-// Cost annotation works without any operator action because every loader
-// starts with this table.
-func DefaultTable() *Table {
-	defaultTableOnce.Do(func() {
-		t, err := parsePricingBytes(defaultPricingYAML)
-		if err != nil {
-			panic(fmt.Sprintf("llmobs: embedded default pricing failed to parse: %v", err))
-		}
-		defaultTablePtr = t
-	})
-	return defaultTablePtr
-}
-
-// mergeOver returns a new Table containing every entry from base, with any
-// matching entry from overlay replacing the base value. Either argument may
-// be nil. Result is a fresh allocation so callers can mutate / Store safely.
-func mergeOver(base, overlay *Table) *Table {
-	if overlay == nil || len(overlay.entries) == 0 {
-		return base
-	}
-	if base == nil || len(base.entries) == 0 {
-		return overlay
-	}
-	out := make(map[string]map[string]Entry, len(base.entries))
-	for provider, models := range base.entries {
-		inner := make(map[string]Entry, len(models))
-		for model, e := range models {
-			inner[model] = e
-		}
-		out[provider] = inner
-	}
-	for provider, models := range overlay.entries {
-		inner, ok := out[provider]
-		if !ok {
-			inner = make(map[string]Entry, len(models))
-			out[provider] = inner
-		}
-		for model, e := range models {
-			inner[model] = e
-		}
-	}
-	return &Table{entries: out}
-}
-
 // Entry is a single model's input and output pricing, expressed in USD per
 // 1000 tokens.
 //
 // CachedInputPer1K applies to OpenAI's cached prompt tokens, which are a
 // subset of input_tokens — when set, the cached portion is billed at this
 // rate and the non-cached remainder at InputPer1K. Zero means "no discount
-// configured", and cached tokens are billed at InputPer1K (matches current
-// behaviour where cached counts weren't extracted at all).
+// configured", and cached tokens are billed at InputPer1K.
 //
 // CacheReadPer1K and CacheCreationPer1K apply to Anthropic's two prompt-
 // cache fields, which are additive to input_tokens: cache_read is the
 // cheaper read-from-cache rate, cache_creation is the more expensive
 // write-to-cache rate. Zero means "no rate configured" and the
-// corresponding token bucket is billed at InputPer1K. This is more
-// accurate than today's behaviour, where Anthropic's cache tokens are
-// ignored and not charged at all.
+// corresponding token bucket is billed at InputPer1K.
 type Entry struct {
 	InputPer1K         float64
 	OutputPer1K        float64
@@ -105,33 +33,102 @@ type Entry struct {
 	CacheCreationPer1K float64
 }

-// Table is a provider-to-model pricing lookup. Instances are immutable once
-// built and are swapped atomically by Loader.
+// EntryJSON is the wire shape of a pricing entry inside cost_meter's
+// middleware config. Field names are the management→proxy contract; the
+// management synthesizer marshals the same names (its pricing.Entry).
+type EntryJSON struct {
+	InputPer1K         float64 `json:"input_per_1k"`
+	OutputPer1K        float64 `json:"output_per_1k"`
+	CachedInputPer1K   float64 `json:"cached_input_per_1k"`
+	CacheReadPer1K     float64 `json:"cache_read_per_1k"`
+	CacheCreationPer1K float64 `json:"cache_creation_per_1k"`
+}
+
+// Table is a provider-surface-to-model pricing lookup. Instances are
+// immutable once built; a mapping update builds a whole new middleware
+// instance (and with it a new table) rather than mutating this one.
 type Table struct {
 	entries map[string]map[string]Entry
 }

+// NewEntries validates and converts a wire-shape map (surface-or-record ->
+// model -> rates) into the internal representation. Every rate must be a
+// finite, non-negative USD amount; a violation is returned as an error so
+// a corrupt config fails the chain build loudly instead of mispricing.
+// Management validates the same constraints at its API boundary, so this
+// is defense-in-depth. Nil input yields an empty (never-matching) map.
+func NewEntries(raw map[string]map[string]EntryJSON) (map[string]map[string]Entry, error) {
+	out := make(map[string]map[string]Entry, len(raw))
+	for outer, models := range raw {
+		inner := make(map[string]Entry, len(models))
+		for model, e := range models {
+			for field, v := range map[string]float64{
+				"input_per_1k":          e.InputPer1K,
+				"output_per_1k":         e.OutputPer1K,
+				"cached_input_per_1k":   e.CachedInputPer1K,
+				"cache_read_per_1k":     e.CacheReadPer1K,
+				"cache_creation_per_1k": e.CacheCreationPer1K,
+			} {
+				if v < 0 || math.IsNaN(v) || math.IsInf(v, 0) {
+					return nil, fmt.Errorf("pricing %s/%s: %s must be a finite, non-negative rate, got %v", outer, model, field, v)
+				}
+			}
+			// EntryJSON and Entry are field-identical (tags aside), so a
+			// direct conversion carries all five rates.
+			inner[model] = Entry(e)
+		}
+		out[outer] = inner
+	}
+	return out, nil
+}
+
+// NewTable builds an immutable Table from the wire-shape defaults map.
+// See NewEntries for validation semantics.
+func NewTable(raw map[string]map[string]EntryJSON) (*Table, error) {
+	entries, err := NewEntries(raw)
+	if err != nil {
+		return nil, err
+	}
+	return &Table{entries: entries}, nil
+}
+
+// Lookup returns the entry for the given provider surface and model.
+func (t *Table) Lookup(provider, model string) (Entry, bool) {
+	if t == nil {
+		return Entry{}, false
+	}
+	byModel, ok := t.entries[provider]
+	if !ok {
+		return Entry{}, false
+	}
+	e, ok := byModel[model]
+	return e, ok
+}
+
+// Has reports whether the provider/model pair is present in the table.
+func (t *Table) Has(provider, model string) bool {
+	_, ok := t.Lookup(provider, model)
+	return ok
+}
+
 // Cost returns the estimated USD cost for the given token counts. ok is
 // false when the provider or model is not present in the table; the caller
 // can still emit token metrics with a model=unknown label.
-//
-// Provider-shape semantics for cached / cache-creation counts:
-//
-//   - OpenAI: cachedInput is a SUBSET of inTokens. The cached portion is
-//     billed at CachedInputPer1K (or InputPer1K when no override), and the
-//     non-cached remainder of inTokens at InputPer1K. cacheCreation is
-//     ignored (OpenAI has no analogue).
-//   - Anthropic: cachedInput (cache_read) and cacheCreation are ADDITIVE to
-//     inTokens. The three buckets are billed at CacheReadPer1K,
-//     CacheCreationPer1K, and InputPer1K respectively, each falling back
-//     to InputPer1K when the corresponding rate is zero.
-//   - Other providers: cached and cacheCreation are ignored; cost is
-//     inTokens*InputPer1K + outTokens*OutputPer1K.
 func (t *Table) Cost(provider, model string, inTokens, outTokens, cachedInput, cacheCreation int64) (float64, bool) {
 	c, ok := t.Costs(provider, model, inTokens, outTokens, cachedInput, cacheCreation)
 	return c.TotalUSD, ok
 }

+// Costs returns the estimated USD cost split for the given token counts.
+// The provider surface selects the cache formula; see EntryCosts.
+func (t *Table) Costs(provider, model string, inTokens, outTokens, cachedInput, cacheCreation int64) (Costs, bool) {
+	entry, ok := t.Lookup(provider, model)
+	if !ok {
+		return Costs{}, false
+	}
+	return EntryCosts(entry, provider, inTokens, outTokens, cachedInput, cacheCreation), true
+}
+
 // Costs is a per-request cost split. The four per-bucket fields are the base
 // of the breakdown — one per token bucket the provider bills separately — and
 // the two aggregates are derived from them:
@@ -165,9 +162,25 @@ func newCosts(input, cachedInput, cacheCreation, output float64) Costs {
 	}
 }

-// Costs returns the estimated USD cost split for the given token counts, with
-// the same semantics as Cost.
-func (t *Table) Costs(provider, model string, inTokens, outTokens, cachedInput, cacheCreation int64) (Costs, bool) {
+// EntryCosts computes the USD cost split for the given entry and token
+// counts. The surface (the llm.provider value the request parser stamps)
+// selects the cache formula; the entry may come from the surface-keyed
+// defaults table or from a per-provider-record override — the math is
+// identical either way.
+//
+// Provider-shape semantics for cached / cache-creation counts:
+//
+//   - "openai": cachedInput is a SUBSET of inTokens. The cached portion is
+//     billed at CachedInputPer1K (or InputPer1K when no override), and the
+//     non-cached remainder of inTokens at InputPer1K. cacheCreation is
+//     ignored (OpenAI has no analogue).
+//   - "anthropic", "bedrock": cachedInput (cache_read) and cacheCreation are
+//     ADDITIVE to inTokens. The three buckets are billed at CacheReadPer1K,
+//     CacheCreationPer1K, and InputPer1K respectively, each falling back
+//     to InputPer1K when the corresponding rate is zero.
+//   - Other surfaces: cached and cacheCreation are ignored; cost is
+//     inTokens*InputPer1K + outTokens*OutputPer1K.
+func EntryCosts(entry Entry, surface string, inTokens, outTokens, cachedInput, cacheCreation int64) Costs {
 	// Clamp negatives to zero before any pricing math so a malformed
 	// upstream count can never produce a negative cost.
 	if inTokens < 0 {
@@ -182,19 +195,8 @@ func (t *Table) Costs(provider, model string, inTokens, outTokens, cachedInput,
 	if cacheCreation < 0 {
 		cacheCreation = 0
 	}
-	if t == nil {
-		return Costs{}, false
-	}
-	byModel, ok := t.entries[provider]
-	if !ok {
-		return Costs{}, false
-	}
-	entry, ok := byModel[model]
-	if !ok {
-		return Costs{}, false
-	}
 	output := (float64(outTokens) / 1000.0) * entry.OutputPer1K
-	switch provider {
+	switch surface {
 	case "openai":
 		// cachedInput is a subset of inTokens; clamp so a malformed
 		// upstream (cached > total) can't produce a negative remainder.
@@ -208,7 +210,7 @@ func (t *Table) Costs(provider, model string, inTokens, outTokens, cachedInput,
 		}
 		nonCached := float64(inTokens-clamped) / 1000.0 * entry.InputPer1K
 		cached := float64(clamped) / 1000.0 * cachedRate
-		return newCosts(nonCached, cached, 0, output), true
+		return newCosts(nonCached, cached, 0, output)
 	case "anthropic", "bedrock":
 		// Bedrock-Anthropic returns the same additive cache buckets as
 		// first-party Anthropic; non-Anthropic Bedrock models simply report
@@ -224,266 +226,9 @@ func (t *Table) Costs(provider, model string, inTokens, outTokens, cachedInput,
 		input := float64(inTokens) / 1000.0 * entry.InputPer1K
 		read := float64(cachedInput) / 1000.0 * readRate
 		create := float64(cacheCreation) / 1000.0 * createRate
-		return newCosts(input, read, create, output), true
+		return newCosts(input, read, create, output)
 	default:
 		input := float64(inTokens) / 1000.0 * entry.InputPer1K
-		return newCosts(input, 0, 0, output), true
+		return newCosts(input, 0, 0, output)
 	}
 }
-
-// Has reports whether the provider/model pair is present in the table.
-func (t *Table) Has(provider, model string) bool {
-	if t == nil {
-		return false
-	}
-	byModel, ok := t.entries[provider]
-	if !ok {
-		return false
-	}
-	_, ok = byModel[model]
-	return ok
-}
-
-// pricingFile mirrors the on-disk YAML schema. Keys are provider names; the
-// nested map keys are model names.
-type pricingFile map[string]map[string]struct {
-	InputPer1K         float64 `yaml:"input_per_1k"`
-	OutputPer1K        float64 `yaml:"output_per_1k"`
-	CachedInputPer1K   float64 `yaml:"cached_input_per_1k"`
-	CacheReadPer1K     float64 `yaml:"cache_read_per_1k"`
-	CacheCreationPer1K float64 `yaml:"cache_creation_per_1k"`
-}
-
-const (
-	// ReloadInterval is the mtime-poll cadence for the background reloader.
-	ReloadInterval = 30 * time.Second
-
-	// errorBackoff bounds how often the loader logs a repeated parse error.
-	errorBackoff = 5 * time.Minute
-)
-
-var basenameRegex = regexp.MustCompile(`^[a-zA-Z0-9._-]+$`)
-
-// Loader is a confined, hot-reloadable pricing table reader. Construction
-// must succeed against the target file; subsequent reload failures keep the
-// previously-loaded table so callers never observe a blank price list.
-type Loader struct {
-	baseDir  string
-	fullPath string
-	pluginID string
-	table    atomic.Pointer[Table]
-	mtime    atomic.Int64
-	failures metric.Int64Counter
-	interval time.Duration
-}
-
-// NewLoader returns a pricing loader that overlays an optional file-based
-// table on top of the embedded defaults. Missing override file, baseDir, or
-// relPath is not an error: the loader keeps the embedded defaults so cost
-// metadata is still emitted for known models.
-//
-// Errors:
-//   - bad basename, traversal segment, or absolute relPath are rejected so a
-//     misconfigured target surfaces immediately.
-//   - permission errors and YAML parse errors keep the defaults but log a
-//     warning; cost annotation does not silently break.
-//
-// failures is optional; pass nil in tests that do not care about
-// reload-failure telemetry.
-func NewLoader(baseDir, relPath, pluginID string, failures metric.Int64Counter) (*Loader, error) {
-	defaults := DefaultTable()
-	l := &Loader{
-		baseDir:  baseDir,
-		pluginID: pluginID,
-		failures: failures,
-	}
-	l.table.Store(defaults)
-
-	if strings.TrimSpace(baseDir) == "" || strings.TrimSpace(relPath) == "" {
-		return l, nil
-	}
-
-	full, err := resolveMiddlewareDataPath(baseDir, relPath)
-	if err != nil {
-		return nil, err
-	}
-	l.fullPath = full
-
-	overlay, mtime, err := loadPricing(full)
-	if err != nil {
-		if errors.Is(err, fs.ErrNotExist) {
-			// Override file is optional. Defaults already stored.
-			return l, nil
-		}
-		// Symlink rejection, oversize file, parse failure, permission errors
-		// — surface so a misconfigured operator sees the problem instead of
-		// silently running with stale defaults.
-		return nil, fmt.Errorf("load pricing %s: %w", full, err)
-	}
-	l.table.Store(mergeOver(defaults, overlay))
-	l.mtime.Store(mtime.UnixNano())
-	return l, nil
-}
-
-// Get returns the current pricing table. The returned pointer is immutable;
-// callers must not mutate its contents.
-func (l *Loader) Get() *Table {
-	if l == nil {
-		return nil
-	}
-	return l.table.Load()
-}
-
-// WatchesFile reports whether this loader is bound to an override file on
-// disk. False for defaults-only loaders (no operator override given).
-// Callers use this to decide whether to spawn the mtime-poll goroutine.
-func (l *Loader) WatchesFile() bool {
-	if l == nil {
-		return false
-	}
-	return l.fullPath != ""
-}
-
-// SetReloadInterval overrides the mtime-poll cadence used by Reload. Calls
-// after Reload has started have no effect on the running loop. Intended for
-// tests; production code uses the default ReloadInterval.
-func (l *Loader) SetReloadInterval(d time.Duration) {
-	if l == nil || d <= 0 {
-		return
-	}
-	l.interval = d
-}
-
-// Reload runs a polling loop that checks the pricing file mtime every
-// ReloadInterval (or the value passed to SetReloadInterval). Returns when
-// ctx is cancelled.
-func (l *Loader) Reload(ctx context.Context) {
-	if l == nil {
-		return
-	}
-	interval := l.interval
-	if interval <= 0 {
-		interval = ReloadInterval
-	}
-	t := time.NewTicker(interval)
-	defer t.Stop()
-
-	var lastErrAt time.Time
-	for {
-		select {
-		case <-ctx.Done():
-			return
-		case <-t.C:
-			if err := l.reload(); err != nil {
-				if l.failures != nil {
-					l.failures.Add(ctx, 1, metric.WithAttributes(
-						attribute.String("plugin", l.pluginID),
-					))
-				}
-				now := time.Now()
-				if now.Sub(lastErrAt) >= errorBackoff {
-					log.Warnf("llmobs: pricing reload failed for %s: %v", l.fullPath, err)
-					lastErrAt = now
-				}
-			}
-		}
-	}
-}
-
-// reload performs a single-shot mtime check and reload. The reloaded
-// override file is merged on top of the embedded defaults; missing override
-// (e.g. operator deleted the file) is not an error and reverts to defaults.
-func (l *Loader) reload() error {
-	if l.fullPath == "" {
-		// Defaults-only loader; nothing on disk to reload.
-		return nil
-	}
-	mtime, err := statMtime(l.fullPath)
-	if err != nil {
-		if errors.Is(err, fs.ErrNotExist) {
-			// File was removed since startup. Drop back to defaults and
-			// reset mtime so a future re-creation triggers a reload.
-			l.table.Store(DefaultTable())
-			l.mtime.Store(0)
-			return nil
-		}
-		return err
-	}
-	if mtime.UnixNano() == l.mtime.Load() {
-		return nil
-	}
-
-	overlay, newMtime, err := loadPricing(l.fullPath)
-	if err != nil {
-		return err
-	}
-	l.table.Store(mergeOver(DefaultTable(), overlay))
-	l.mtime.Store(newMtime.UnixNano())
-	return nil
-}
-
-// resolveMiddlewareDataPath validates relPath is a safe basename and resolves
-// it under baseDir. An additional cleaned-prefix check guards against
-// CVE-style edge cases where Join is used with trailing path segments.
-func resolveMiddlewareDataPath(baseDir, relPath string) (string, error) {
-	if strings.TrimSpace(baseDir) == "" {
-		return "", errors.New("middleware-data-dir is not configured")
-	}
-	if relPath == "" {
-		return "", errors.New("pricing path is empty")
-	}
-	if !basenameRegex.MatchString(relPath) {
-		return "", fmt.Errorf("pricing path %q is not a safe basename", relPath)
-	}
-	if filepath.IsAbs(relPath) {
-		return "", fmt.Errorf("pricing path %q must be a basename, not absolute", relPath)
-	}
-
-	cleanBase, err := filepath.Abs(filepath.Clean(baseDir))
-	if err != nil {
-		return "", fmt.Errorf("resolve middleware-data-dir: %w", err)
-	}
-	full := filepath.Join(cleanBase, relPath)
-	cleanedFull := filepath.Clean(full)
-	if !strings.HasPrefix(cleanedFull, cleanBase+string(filepath.Separator)) && cleanedFull != cleanBase {
-		return "", fmt.Errorf("pricing path %q escapes middleware-data-dir", relPath)
-	}
-	return cleanedFull, nil
-}
-
-func parsePricingBytes(data []byte) (*Table, error) {
-	dec := yaml.NewDecoder(bytes.NewReader(data))
-	dec.KnownFields(true)
-
-	var raw pricingFile
-	if err := dec.Decode(&raw); err != nil && !errors.Is(err, io.EOF) {
-		return nil, fmt.Errorf("decode pricing yaml: %w", err)
-	}
-
-	out := make(map[string]map[string]Entry, len(raw))
-	for provider, models := range raw {
-		inner := make(map[string]Entry, len(models))
-		for model, entry := range models {
-			for field, v := range map[string]float64{
-				"input_per_1k":          entry.InputPer1K,
-				"output_per_1k":         entry.OutputPer1K,
-				"cached_input_per_1k":   entry.CachedInputPer1K,
-				"cache_read_per_1k":     entry.CacheReadPer1K,
-				"cache_creation_per_1k": entry.CacheCreationPer1K,
-			} {
-				if v < 0 || math.IsNaN(v) || math.IsInf(v, 0) {
-					return nil, fmt.Errorf("pricing %s/%s: %s must be a finite, non-negative rate, got %v", provider, model, field, v)
-				}
-			}
-			inner[model] = Entry{
-				InputPer1K:         entry.InputPer1K,
-				OutputPer1K:        entry.OutputPer1K,
-				CachedInputPer1K:   entry.CachedInputPer1K,
-				CacheReadPer1K:     entry.CacheReadPer1K,
-				CacheCreationPer1K: entry.CacheCreationPer1K,
-			}
-		}
-		out[provider] = inner
-	}
-	return &Table{entries: out}, nil
-}
--- a/proxy/internal/llm/pricing/pricing_other.go
+++ b/proxy/internal/llm/pricing/pricing_other.go
@@ -1,20 +0,0 @@
-//go:build !unix
-
-package pricing
-
-import (
-	"fmt"
-	"time"
-)
-
-// loadPricing is unavailable on non-Unix platforms because O_NOFOLLOW and
-// fstat-from-FD are required to honour the spec's symlink-safety rules. The
-// proxy is only deployed on Linux today; a Windows port would need an
-// equivalent path-as-handle implementation.
-func loadPricing(path string) (*Table, time.Time, error) {
-	return nil, time.Time{}, fmt.Errorf("llmobs pricing loader is not supported on this platform: %s", path)
-}
-
-func statMtime(path string) (time.Time, error) {
-	return time.Time{}, fmt.Errorf("llmobs pricing loader is not supported on this platform: %s", path)
-}
--- a/proxy/internal/llm/pricing/pricing_test.go
+++ b/proxy/internal/llm/pricing/pricing_test.go
@@ -1,47 +1,13 @@
-//go:build unix
-
 package pricing

 import (
-	"context"
-	"os"
-	"path/filepath"
+	"math"
 	"testing"
-	"time"

 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )

-func copyFixture(t *testing.T, src, dst string) {
-	t.Helper()
-	data, err := os.ReadFile(src)
-	require.NoError(t, err, "read source fixture")
-	require.NoError(t, os.WriteFile(dst, data, 0o600), "write target fixture")
-}
-
-func TestNewLoader_HappyPath(t *testing.T) {
-	base := t.TempDir()
-	copyFixture(t, filepath.Join("..", "fixtures", "pricing.yaml"), filepath.Join(base, "pricing.yaml"))
-
-	l, err := NewLoader(base, "pricing.yaml", "llm_observability", nil)
-	require.NoError(t, err, "NewLoader must succeed with a valid fixture")
-	table := l.Get()
-	require.NotNil(t, table, "table populated after load")
-
-	cost, ok := table.Cost("openai", "gpt-4o-mini", 1000, 1000, 0, 0)
-	require.True(t, ok, "known provider/model resolves")
-	assert.InDelta(t, 0.00075, cost, 1e-9, "cost = 0.00015 + 0.0006 per 1k tokens")
-
-	cost, ok = table.Cost("openai", "gpt-4o", 2000, 1000, 0, 0)
-	require.True(t, ok, "second known model resolves")
-	assert.InDelta(t, 0.015, cost, 1e-9, "cost for gpt-4o: 2*0.0025 + 1*0.01")
-
-	cost, ok = table.Cost("anthropic", "claude-sonnet-4-5", 1000, 1000, 0, 0)
-	require.True(t, ok, "anthropic model resolves")
-	assert.InDelta(t, 0.018, cost, 1e-9, "cost for claude-sonnet-4-5: 0.003 + 0.015")
-}
-
 // TestCost_OpenAICachedSubsetDiscount proves OpenAI's cached input
 // tokens are billed at the configured cached_input_per_1k rate while
 // the non-cached remainder of input_tokens is billed at the regular
@@ -65,11 +31,9 @@ func TestCost_OpenAICachedSubsetDiscount(t *testing.T) {
 		"cached subset must bill at the discount rate; non-cached remainder at regular rate")
 }

-// TestCost_OpenAICachedFallsBackToInputRate covers the operator
-// opt-in contract: when CachedInputPer1K is unset (zero), cached
-// tokens bill at the regular input rate. This matches today's
-// behaviour (cached counts weren't extracted at all so they
-// implicitly billed at the input rate via prompt_tokens).
+// TestCost_OpenAICachedFallsBackToInputRate covers the fallback
+// contract: when CachedInputPer1K is unset (zero), cached tokens bill
+// at the regular input rate.
 func TestCost_OpenAICachedFallsBackToInputRate(t *testing.T) {
 	tbl := &Table{entries: map[string]map[string]Entry{
 		"openai": {"gpt-4o": {InputPer1K: 0.0025, OutputPer1K: 0.01}},
@@ -78,7 +42,7 @@ func TestCost_OpenAICachedFallsBackToInputRate(t *testing.T) {
 	require.True(t, ok)
 	want := 0.0025 + (500.0/1000.0)*0.01
 	assert.InDelta(t, want, cost, 1e-12,
-		"absent cached_input_per_1k rate must fall back to input_per_1k — same as pre-feature behaviour")
+		"absent cached_input_per_1k rate must fall back to input_per_1k")
 }

 // TestCost_OpenAIClampsCachedToInputCount is the defensive guard
@@ -100,10 +64,7 @@ func TestCost_OpenAIClampsCachedToInputCount(t *testing.T) {
 // TestCost_AnthropicCacheReadAndCreationAreAdditive proves the
 // Anthropic shape: cache_read and cache_creation tokens are
 // ADDITIVE to input_tokens (not subset), each billed at its own
-// configured rate. The two rates pull in opposite directions —
-// cache_read is the cheaper read-from-cache rate (≈0.1× input),
-// cache_creation is the more expensive write-to-cache rate
-// (≈1.25× input).
+// configured rate.
 func TestCost_AnthropicCacheReadAndCreationAreAdditive(t *testing.T) {
 	tbl := &Table{entries: map[string]map[string]Entry{
 		"anthropic": {"claude-sonnet": {
@@ -125,11 +86,9 @@ func TestCost_AnthropicCacheReadAndCreationAreAdditive(t *testing.T) {
 		"each Anthropic input bucket must bill at its own configured rate")
 }

-// TestCost_AnthropicCacheRatesFallBackToInput covers the no-opt-in
+// TestCost_AnthropicCacheRatesFallBackToInput covers the no-rate
 // path: when neither CacheReadPer1K nor CacheCreationPer1K is set,
-// cache tokens bill at the regular input rate. This is more
-// accurate than today's behaviour (cache tokens ignored entirely)
-// without requiring operators to opt in via YAML.
+// cache tokens bill at the regular input rate.
 func TestCost_AnthropicCacheRatesFallBackToInput(t *testing.T) {
 	tbl := &Table{entries: map[string]map[string]Entry{
 		"anthropic": {"claude-sonnet": {InputPer1K: 0.003, OutputPer1K: 0.015}},
@@ -139,259 +98,39 @@ func TestCost_AnthropicCacheRatesFallBackToInput(t *testing.T) {
 	// Without overrides: every input bucket at input_per_1k.
 	want := ((256.0+768.0+512.0)/1000.0)*0.003 + (200.0/1000.0)*0.015
 	assert.InDelta(t, want, cost, 1e-12,
-		"absent cache rates must fall back to input_per_1k — Anthropic cache tokens were ignored before this change, billing at input rate is more accurate as a default")
+		"absent cache rates must fall back to input_per_1k")
 }

-func TestNewLoader_UnknownModel(t *testing.T) {
-	base := t.TempDir()
-	copyFixture(t, filepath.Join("..", "fixtures", "pricing.yaml"), filepath.Join(base, "pricing.yaml"))
+// TestEntryCosts_SurfaceSelectsFormula pins that the formula branches on
+// the SURFACE, not on which table the entry came from: the same entry
+// bills a subset carve-out on "openai", additive buckets on
+// "anthropic"/"bedrock", and ignores cache counts everywhere else. This
+// is what keeps per-provider-record entries (looked up by record id)
+// mathematically identical to defaults-table entries.
+func TestEntryCosts_SurfaceSelectsFormula(t *testing.T) {
+	e := Entry{InputPer1K: 0.002, OutputPer1K: 0.01, CachedInputPer1K: 0.001, CacheReadPer1K: 0.0002, CacheCreationPer1K: 0.0025}

-	l, err := NewLoader(base, "pricing.yaml", "llm_observability", nil)
-	require.NoError(t, err)
+	openai := EntryCosts(e, "openai", 1000, 0, 400, 300)
+	assert.InDelta(t, (600.0/1000.0)*0.002+(400.0/1000.0)*0.001, openai.TotalUSD, 1e-12,
+		"openai: cached is a subset, cacheCreation ignored")

-	_, ok := l.Get().Cost("openai", "fantasy-model", 10, 10, 0, 0)
-	assert.False(t, ok, "unknown model returns ok=false")
+	anthropic := EntryCosts(e, "anthropic", 1000, 0, 400, 300)
+	assert.InDelta(t, 0.002+(400.0/1000.0)*0.0002+(300.0/1000.0)*0.0025, anthropic.TotalUSD, 1e-12,
+		"anthropic: cache buckets are additive")

-	_, ok = l.Get().Cost("cohere", "anything", 10, 10, 0, 0)
-	assert.False(t, ok, "unknown provider returns ok=false")
+	bedrock := EntryCosts(e, "bedrock", 1000, 0, 400, 300)
+	assert.InDelta(t, anthropic.TotalUSD, bedrock.TotalUSD, 1e-12, "bedrock shares the anthropic formula")
+
+	other := EntryCosts(e, "gemini", 1000, 0, 400, 300)
+	assert.InDelta(t, 0.002, other.TotalUSD, 1e-12, "unknown surface: cache counts ignored")
 }

-func TestNewLoader_InvalidYAMLRejected(t *testing.T) {
-	base := t.TempDir()
-	require.NoError(t, os.WriteFile(filepath.Join(base, "pricing.yaml"), []byte("\t- this is not: valid: yaml: :["), 0o600))
-
-	_, err := NewLoader(base, "pricing.yaml", "llm_observability", nil)
-	require.Error(t, err, "invalid YAML must surface as construction error")
-}
-
-func TestLoader_ReloadKeepsPreviousOnParseError(t *testing.T) {
-	base := t.TempDir()
-	target := filepath.Join(base, "pricing.yaml")
-	copyFixture(t, filepath.Join("..", "fixtures", "pricing.yaml"), target)
-
-	l, err := NewLoader(base, "pricing.yaml", "llm_observability", nil)
-	require.NoError(t, err)
-	require.NotNil(t, l.Get(), "initial table populated")
-
-	// Overwrite with content that violates the strict schema (extra field)
-	// plus a bumped mtime to trigger reload.
-	require.NoError(t, os.WriteFile(target, []byte("openai:\n  gpt-4o:\n    input_per_1k: 1.0\n    output_per_1k: 2.0\n    bogus_field: nope\n"), 0o600))
-	future := time.Now().Add(time.Hour)
-	require.NoError(t, os.Chtimes(target, future, future))
-
-	err = l.reload()
-	require.Error(t, err, "parse error surfaced by reload()")
-
-	cost, ok := l.Get().Cost("openai", "gpt-4o-mini", 1000, 1000, 0, 0)
-	require.True(t, ok, "previous table still available after parse failure")
-	assert.InDelta(t, 0.00075, cost, 1e-9, "previous cost preserved")
-}
-
-func TestLoader_ReloadNoChangeIsNoOp(t *testing.T) {
-	base := t.TempDir()
-	target := filepath.Join(base, "pricing.yaml")
-	copyFixture(t, filepath.Join("..", "fixtures", "pricing.yaml"), target)
-
-	l, err := NewLoader(base, "pricing.yaml", "llm_observability", nil)
-	require.NoError(t, err)
-	ptrBefore := l.Get()
-
-	require.NoError(t, l.reload(), "no-change reload must not error")
-	ptrAfter := l.Get()
-	assert.Same(t, ptrBefore, ptrAfter, "table pointer unchanged when mtime unchanged")
-}
-
-func TestLoader_ReloadDetectsChange(t *testing.T) {
-	base := t.TempDir()
-	target := filepath.Join(base, "pricing.yaml")
-	copyFixture(t, filepath.Join("..", "fixtures", "pricing.yaml"), target)
-
-	l, err := NewLoader(base, "pricing.yaml", "llm_observability", nil)
-	require.NoError(t, err)
-
-	updated := []byte("openai:\n  gpt-4o-mini:\n    input_per_1k: 1.00\n    output_per_1k: 2.00\n")
-	require.NoError(t, os.WriteFile(target, updated, 0o600))
-	future := time.Now().Add(time.Hour)
-	require.NoError(t, os.Chtimes(target, future, future))
-
-	require.NoError(t, l.reload(), "reload must succeed on valid new content")
-
-	cost, ok := l.Get().Cost("openai", "gpt-4o-mini", 1000, 1000, 0, 0)
-	require.True(t, ok, "updated model still present")
-	assert.InDelta(t, 3.0, cost, 0.0001, "new prices are applied: 1 + 2 per 1k")
-}
-
-// TestLoader_ReloadGoroutinePicksUpChanges proves the background goroutine
-// started via Reload actually swaps the pricing table when the file changes
-// on disk. Without that goroutine running, pricing edits would never reach
-// requests until a proxy restart.
-func TestLoader_ReloadGoroutinePicksUpChanges(t *testing.T) {
-	base := t.TempDir()
-	target := filepath.Join(base, "pricing.yaml")
-	copyFixture(t, filepath.Join("..", "fixtures", "pricing.yaml"), target)
-
-	l, err := NewLoader(base, "pricing.yaml", "llm_observability", nil)
-	require.NoError(t, err)
-	l.SetReloadInterval(20 * time.Millisecond)
-
-	ctx, cancel := context.WithCancel(context.Background())
-	defer cancel()
-	done := make(chan struct{})
-	go func() {
-		l.Reload(ctx)
-		close(done)
-	}()
-
-	// Before any rewrite, the loader holds the fixture's prices.
-	costBefore, ok := l.Get().Cost("openai", "gpt-4o-mini", 1000, 1000, 0, 0)
-	require.True(t, ok, "fixture model must resolve initially")
-	assert.InDelta(t, 0.00075, costBefore, 1e-9, "fixture prices apply before rewrite")
-
-	updated := []byte("openai:\n  gpt-4o-mini:\n    input_per_1k: 1.00\n    output_per_1k: 2.00\n")
-	require.NoError(t, os.WriteFile(target, updated, 0o600))
-	future := time.Now().Add(time.Hour)
-	require.NoError(t, os.Chtimes(target, future, future))
-
-	deadline := time.Now().Add(2 * time.Second)
-	for {
-		cost, ok := l.Get().Cost("openai", "gpt-4o-mini", 1000, 1000, 0, 0)
-		if ok && cost > 2.5 {
-			break
-		}
-		if time.Now().After(deadline) {
-			t.Fatalf("background reloader did not pick up rewrite within deadline")
-		}
-		time.Sleep(10 * time.Millisecond)
-	}
-
-	cancel()
-	select {
-	case <-done:
-	case <-time.After(2 * time.Second):
-		t.Fatal("Reload loop did not exit after cancel")
-	}
-}
-
-func TestLoader_ReloadBackgroundLoopCancellation(t *testing.T) {
-	base := t.TempDir()
-	copyFixture(t, filepath.Join("..", "fixtures", "pricing.yaml"), filepath.Join(base, "pricing.yaml"))
-	l, err := NewLoader(base, "pricing.yaml", "llm_observability", nil)
-	require.NoError(t, err)
-
-	ctx, cancel := context.WithCancel(context.Background())
-	done := make(chan struct{})
-	go func() {
-		l.Reload(ctx)
-		close(done)
-	}()
-	cancel()
-
-	select {
-	case <-done:
-	case <-time.After(2 * time.Second):
-		t.Fatal("Reload loop did not exit on context cancel")
-	}
-}
-
-func TestNewLoader_PathValidation(t *testing.T) {
-	base := t.TempDir()
-	copyFixture(t, filepath.Join("..", "fixtures", "pricing.yaml"), filepath.Join(base, "pricing.yaml"))
-
-	cases := []struct {
-		name    string
-		relPath string
-	}{
-		{"traversal", "../../etc/passwd"},
-		{"absolute", "/etc/passwd"},
-		{"slash in basename", "sub/pricing.yaml"},
-		{"control chars", "pricing\x00.yaml"},
-	}
-	for _, tc := range cases {
-		t.Run(tc.name, func(t *testing.T) {
-			_, err := NewLoader(base, tc.relPath, "llm_observability", nil)
-			require.Error(t, err, "NewLoader must reject %q", tc.relPath)
-		})
-	}
-
-	// Empty relPath is no longer a validation error: the loader treats it
-	// as "no override file, defaults only" so cost metadata is still
-	// emitted for the embedded models out of the box.
-	t.Run("empty falls back to defaults", func(t *testing.T) {
-		l, err := NewLoader(base, "", "llm_observability", nil)
-		require.NoError(t, err, "empty relPath should yield a defaults-only loader")
-		require.NotNil(t, l, "loader must be returned")
-		require.False(t, l.WatchesFile(), "no file watching when no override is given")
-		_, ok := l.Get().Cost("openai", "gpt-4o-mini", 1000, 1000, 0, 0)
-		assert.True(t, ok, "embedded defaults should still resolve gpt-4o-mini")
-	})
-}
-
-// TestNewLoader_PathValidation_Extended covers the remaining attack shapes
-// called out in C2: dot references, embedded traversal segments, and a
-// newline in the basename. The basename regex must reject each one even
-// though filepath.Clean would otherwise collapse them.
-func TestNewLoader_PathValidation_Extended(t *testing.T) {
-	base := t.TempDir()
-	copyFixture(t, filepath.Join("..", "fixtures", "pricing.yaml"), filepath.Join(base, "pricing.yaml"))
-
-	cases := []struct {
-		name    string
-		relPath string
-	}{
-		{"dot", "."},
-		{"dotdot", ".."},
-		{"relative traversal", "../pricing.yaml"},
-		{"embedded slash", "pri/cing.yaml"},
-		{"newline", "pricing\n.yaml"},
-	}
-	for _, tc := range cases {
-		t.Run(tc.name, func(t *testing.T) {
-			_, err := NewLoader(base, tc.relPath, "llm_observability", nil)
-			require.Error(t, err, "NewLoader must reject %q", tc.relPath)
-		})
-	}
-}
-
-// TestNewLoader_ValidBasenameLoads proves the allowlist is exclusive: a
-// basename containing only safe characters under baseDir loads. Without this
-// a regression that over-tightened the regex would silently break valid
-// deployments.
-func TestNewLoader_ValidBasenameLoads(t *testing.T) {
-	base := t.TempDir()
-	copyFixture(t, filepath.Join("..", "fixtures", "pricing.yaml"), filepath.Join(base, "pricing-v2_prod.yaml"))
-
-	l, err := NewLoader(base, "pricing-v2_prod.yaml", "llm_observability", nil)
-	require.NoError(t, err, "basename with _, -, . must load")
-	require.NotNil(t, l.Get(), "table populated")
-}
-
-// TestNewLoader_SymlinkOutsideBaseDirRejected constructs a symlink under
-// baseDir that points to a file outside it. O_NOFOLLOW must refuse to open
-// the symlink even though the symlink path itself is a valid basename under
-// baseDir.
-func TestNewLoader_SymlinkOutsideBaseDirRejected(t *testing.T) {
-	outside := t.TempDir()
-	target := filepath.Join(outside, "evil.yaml")
-	copyFixture(t, filepath.Join("..", "fixtures", "pricing.yaml"), target)
-
-	base := t.TempDir()
-	link := filepath.Join(base, "pricing.yaml")
-	require.NoError(t, os.Symlink(target, link), "symlink setup")
-
-	_, err := NewLoader(base, "pricing.yaml", "llm_observability", nil)
-	require.Error(t, err, "O_NOFOLLOW must reject symlink even when it points outside baseDir")
-}
-
-func TestNewLoader_SymlinkRejected(t *testing.T) {
-	base := t.TempDir()
-	concrete := filepath.Join(base, "real.yaml")
-	copyFixture(t, filepath.Join("..", "fixtures", "pricing.yaml"), concrete)
-
-	link := filepath.Join(base, "pricing.yaml")
-	require.NoError(t, os.Symlink(concrete, link), "symlink setup")
-
-	_, err := NewLoader(base, "pricing.yaml", "llm_observability", nil)
-	require.Error(t, err, "O_NOFOLLOW must reject symlinked targets")
+// TestEntryCosts_ClampsNegativeTokens: malformed upstream counts must
+// never produce a negative cost.
+func TestEntryCosts_ClampsNegativeTokens(t *testing.T) {
+	e := Entry{InputPer1K: 0.002, OutputPer1K: 0.01}
+	c := EntryCosts(e, "openai", -50, -10, -5, -3)
+	assert.Zero(t, c.TotalUSD, "all-negative counts clamp to zero cost")
 }

 func TestTableCost_NilSafe(t *testing.T) {
@@ -402,31 +141,37 @@ func TestTableCost_NilSafe(t *testing.T) {
 	assert.False(t, t1.Has("x", "y"), "nil table has nothing")
 }

-func TestLoaderGet_NilSafe(t *testing.T) {
-	var l *Loader
-	assert.Nil(t, l.Get(), "nil loader returns nil table")
-}
-
-// TestNewLoader_RejectsOversizedFile_FixesM4 proves the loader bounds reads
-// at maxPricingBytes so a hostile file cannot exhaust process memory.
-func TestNewLoader_RejectsOversizedFile_FixesM4(t *testing.T) {
-	base := t.TempDir()
-	target := filepath.Join(base, "pricing.yaml")
-
-	// Build a YAML payload larger than the cap. We pad with valid YAML
-	// comments so a partial read would still fail the size check rather
-	// than the parser.
-	header := "openai:\n"
-	bigComment := make([]byte, maxPricingBytes+1024)
-	for i := range bigComment {
-		bigComment[i] = ' '
+func TestNewTable_ValidatesRates(t *testing.T) {
+	good := map[string]map[string]EntryJSON{
+		"openai": {"gpt-4o": {InputPer1K: 0.0025, OutputPer1K: 0.01, CachedInputPer1K: 0.00125}},
 	}
-	bigComment[0] = '#'
-	bigComment[len(bigComment)-1] = '\n'
-	payload := append([]byte(header), bigComment...)
-	require.NoError(t, os.WriteFile(target, payload, 0o600))
+	tbl, err := NewTable(good)
+	require.NoError(t, err)
+	cost, ok := tbl.Cost("openai", "gpt-4o", 1000, 1000, 0, 0)
+	require.True(t, ok, "entry survives the wire conversion")
+	assert.InDelta(t, 0.0125, cost, 1e-9)

-	_, err := NewLoader(base, "pricing.yaml", "llm_observability", nil)
-	require.Error(t, err, "oversized pricing file must be rejected")
-	assert.Contains(t, err.Error(), "exceeds", "rejection must reference the byte cap")
+	_, ok = tbl.Cost("openai", "unknown-model", 1, 1, 0, 0)
+	assert.False(t, ok, "unknown model misses")
+
+	for name, bad := range map[string]EntryJSON{
+		"negative input":  {InputPer1K: -1, OutputPer1K: 0.01},
+		"NaN output":      {InputPer1K: 0.01, OutputPer1K: math.NaN()},
+		"Inf cache read":  {InputPer1K: 0.01, OutputPer1K: 0.01, CacheReadPer1K: math.Inf(1)},
+		"negative cached": {InputPer1K: 0.01, OutputPer1K: 0.01, CachedInputPer1K: -0.001},
+	} {
+		_, err := NewTable(map[string]map[string]EntryJSON{"openai": {"m": bad}})
+		assert.Error(t, err, "case %q must be rejected so a corrupt config fails the chain build instead of mispricing", name)
+	}
+}
+
+func TestNewTable_NilAndEmpty(t *testing.T) {
+	tbl, err := NewTable(nil)
+	require.NoError(t, err, "nil map builds an empty (never-matching) table")
+	_, ok := tbl.Cost("openai", "gpt-4o", 1, 1, 0, 0)
+	assert.False(t, ok, "empty table prices nothing")
+
+	entries, err := NewEntries(nil)
+	require.NoError(t, err)
+	assert.Empty(t, entries, "nil in, empty (never-matching) map out for the per-record map")
 }
--- a/proxy/internal/llm/pricing/pricing_unix.go
+++ b/proxy/internal/llm/pricing/pricing_unix.go
@@ -1,68 +0,0 @@
-//go:build unix
-
-package pricing
-
-import (
-	"fmt"
-	"io"
-	"os"
-	"syscall"
-	"time"
-
-	log "github.com/sirupsen/logrus"
-)
-
-// maxPricingBytes caps the size of the pricing YAML on read so a hostile or
-// runaway file cannot exhaust process memory during reload. 1 MiB is several
-// orders of magnitude larger than any reasonable pricing table.
-const maxPricingBytes int64 = 1 << 20
-
-// loadPricing opens the file with O_NOFOLLOW, fstats the open descriptor,
-// and parses from that same descriptor. Never re-opens by path so a
-// mid-read rename or symlink swap cannot substitute content. Bytes are
-// capped at maxPricingBytes so the loader cannot be coerced into reading an
-// unbounded file.
-func loadPricing(path string) (*Table, time.Time, error) {
-	f, err := os.OpenFile(path, os.O_RDONLY|syscall.O_NOFOLLOW, 0)
-	if err != nil {
-		return nil, time.Time{}, fmt.Errorf("open %s: %w", path, err)
-	}
-	defer func() {
-		if cerr := f.Close(); cerr != nil {
-			log.Debugf("close pricing file %s: %v", path, cerr)
-		}
-	}()
-
-	info, err := f.Stat()
-	if err != nil {
-		return nil, time.Time{}, fmt.Errorf("fstat %s: %w", path, err)
-	}
-	if !info.Mode().IsRegular() {
-		return nil, time.Time{}, fmt.Errorf("pricing file %s is not a regular file", path)
-	}
-
-	data, err := io.ReadAll(io.LimitReader(f, maxPricingBytes+1))
-	if err != nil {
-		return nil, time.Time{}, fmt.Errorf("read %s: %w", path, err)
-	}
-	if int64(len(data)) > maxPricingBytes {
-		return nil, time.Time{}, fmt.Errorf("pricing file %s exceeds %d bytes", path, maxPricingBytes)
-	}
-
-	table, err := parsePricingBytes(data)
-	if err != nil {
-		return nil, time.Time{}, err
-	}
-	return table, info.ModTime(), nil
-}
-
-// statMtime returns the mtime of the file at path. It uses lstat semantics
-// via os.Lstat so a symlink swap is detected even though O_NOFOLLOW will
-// later reject the open.
-func statMtime(path string) (time.Time, error) {
-	info, err := os.Lstat(path)
-	if err != nil {
-		return time.Time{}, fmt.Errorf("lstat %s: %w", path, err)
-	}
-	return info.ModTime(), nil
-}
--- a/proxy/internal/middleware/builtin/builtin.go
+++ b/proxy/internal/middleware/builtin/builtin.go
@@ -36,15 +36,13 @@ var defaultRegistry = middleware.NewRegistry()

 // FactoryContext is the per-process bag that concrete factories may
 // consult during construction. It carries the proxy-lifetime context,
-// the data directory used for static config files (pricing tables,
-// allowlists), the OTel meter, and the proxy logger.
+// the OTel meter, and the proxy logger.
 //
 // Configure must be called once at boot before any chain build calls
 // Resolve. Calling it twice overwrites the prior value; tests may rely
 // on this to reset state.
 type FactoryContext struct {
 	Context    context.Context
-	DataDir    string
 	Meter      metric.Meter
 	Logger     *log.Logger
 	MgmtClient MgmtClient
@@ -58,12 +56,11 @@ var (
 // Configure stores the per-process FactoryContext. Concrete factories
 // reach for it via Context(). mgmt may be nil on tests / standalone
 // builds with no management server; consumers must guard.
-func Configure(ctx context.Context, dataDir string, meter metric.Meter, logger *log.Logger, mgmt MgmtClient) {
+func Configure(ctx context.Context, meter metric.Meter, logger *log.Logger, mgmt MgmtClient) {
 	ctxMu.Lock()
 	defer ctxMu.Unlock()
 	ctxStore = FactoryContext{
 		Context:    ctx,
-		DataDir:    dataDir,
 		Meter:      meter,
 		Logger:     logger,
 		MgmtClient: mgmt,
--- a/proxy/internal/middleware/builtin/cost_calculation_matrix_test.go
+++ b/proxy/internal/middleware/builtin/cost_calculation_matrix_test.go
@@ -12,6 +12,7 @@ import (
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"

+	mgmtpricing "github.com/netbirdio/netbird/management/internals/modules/agentnetwork/pricing"
 	"github.com/netbirdio/netbird/proxy/internal/middleware"
 	"github.com/netbirdio/netbird/proxy/internal/middleware/builtin"
 	"github.com/netbirdio/netbird/proxy/internal/middleware/builtin/cost_meter"
@@ -19,17 +20,20 @@ import (
 	"github.com/netbirdio/netbird/proxy/internal/middleware/builtin/llm_response_parser"
 )

-// Drives the real pipeline (llm_request_parser → llm_response_parser → cost_meter) on the embedded default pricing
-// table and asserts exact USD amounts hardcoded from the vendors' published prices, including the cache split.
+// Drives the real pipeline (llm_request_parser → llm_response_parser → cost_meter) on the REAL default pricing
+// table management ships (mgmtpricing.DefaultTable, catalog-derived) and asserts exact USD amounts hardcoded from
+// the vendors' published prices, including the cache split. This is the cross-stack pricing contract test: the
+// management-side Entry JSON must decode into the proxy-side table and produce these exact costs.
 func TestCostCalculation_ProviderMatrix(t *testing.T) {
-	// Empty data dir → embedded defaults, like a proxy with no pricing override.
-	builtin.Configure(context.Background(), t.TempDir(), nil, nil, nil)
+	builtin.Configure(context.Background(), nil, nil, nil)

 	reqMW, err := llm_request_parser.Factory{}.New(nil)
 	require.NoError(t, err, "build llm_request_parser")
 	respMW, err := llm_response_parser.Factory{}.New(nil)
 	require.NoError(t, err, "build llm_response_parser")
-	costMW, err := cost_meter.Factory{}.New(nil)
+	costCfgJSON, err := json.Marshal(map[string]any{"pricing": map[string]any{"defaults": mgmtpricing.DefaultTable()}})
+	require.NoError(t, err, "marshal management default table into cost_meter config")
+	costMW, err := cost_meter.Factory{}.New(costCfgJSON)
 	require.NoError(t, err, "build cost_meter")
 	t.Cleanup(func() { _ = costMW.Close() })

--- a/proxy/internal/middleware/builtin/cost_meter/factory.go
+++ b/proxy/internal/middleware/builtin/cost_meter/factory.go
@@ -2,7 +2,6 @@ package cost_meter

 import (
 	"bytes"
-	"context"
 	"encoding/json"
 	"fmt"

@@ -11,16 +10,27 @@ import (
 	"github.com/netbirdio/netbird/proxy/internal/middleware/builtin"
 )

-// defaultPricingFilename is the basename probed inside the proxy data
-// directory when no override is configured.
-const defaultPricingFilename = "pricing.yaml"
-
-// Config is the on-wire configuration for the middleware.
+// Config is the on-wire configuration for the middleware, synthesized by
+// management (buildCostMeterConfigJSON). The proxy has no embedded price
+// list: this payload is the only pricing source, and updates arrive as
+// ordinary mapping pushes that rebuild the chain (and with it this
+// middleware instance) — no per-request fetches, no reload loops.
 type Config struct {
-	// PricingPath optionally overrides the basename of the pricing
-	// file probed inside the proxy data directory. When empty the
-	// loader falls back to "pricing.yaml".
-	PricingPath string `json:"pricing_path"`
+	Pricing *PricingConfig `json:"pricing"`
+}
+
+// PricingConfig carries the full pricing table:
+//   - Defaults: parser surface ("openai"/"anthropic"/"bedrock") ->
+//     normalized model id -> rates, matched against llm.provider +
+//     llm.model.
+//   - Providers: provider record id -> normalized model id -> rates,
+//     matched against the llm.resolved_provider_id metadata llm_router
+//     stamps. Entries arrive fully materialized (management folds default
+//     cache rates in at synth time), so lookup order is simply
+//     per-record first, defaults second.
+type PricingConfig struct {
+	Defaults  map[string]map[string]pricing.EntryJSON `json:"defaults"`
+	Providers map[string]map[string]pricing.EntryJSON `json:"providers"`
 }

 // Factory builds cost_meter instances from raw config bytes.
@@ -29,45 +39,45 @@ type Factory struct{}
 // ID returns the registry identifier.
 func (Factory) ID() string { return ID }

-// New constructs a middleware instance. Empty, null, and {} configs
-// are accepted; non-empty rawConfig that fails to unmarshal is
-// rejected so misconfigurations surface at chain build time. The
-// pricing loader is built once per instance and reused across
-// invocations.
+// New constructs a middleware instance. Empty, null, and {} configs are
+// accepted for backward compatibility with a management server that
+// predates config-delivered pricing — the instance then skips every cost
+// computation (unknown_model) and a warning is logged once at build time.
+// Non-empty rawConfig that fails to unmarshal, or a table carrying a
+// non-finite / negative rate, is rejected so misconfigurations surface at
+// chain build time.
 func (Factory) New(rawConfig []byte) (middleware.Middleware, error) {
 	cfg, err := decodeConfig(rawConfig)
 	if err != nil {
 		return nil, err
 	}

-	fctx := builtin.Context()
-	pricingPath := cfg.PricingPath
-	if pricingPath == "" {
-		pricingPath = defaultPricingFilename
+	if cfg.Pricing == nil {
+		if logger := builtin.Context().Logger; logger != nil {
+			logger.Warnf("cost_meter: no pricing table in middleware config; management predates config-delivered pricing — every request will record cost.skipped=unknown_model ($0)")
+		}
+		return newMiddleware(mustEmptyTable(), nil), nil
 	}

-	loader, err := pricing.NewLoader(fctx.DataDir, pricingPath, ID, nil)
+	defaults, err := pricing.NewTable(cfg.Pricing.Defaults)
 	if err != nil {
-		return nil, fmt.Errorf("init pricing loader: %w", err)
+		return nil, fmt.Errorf("cost_meter pricing defaults: %w", err)
 	}
-
-	cancel := startReloader(fctx.Context, loader)
-
-	return newMiddleware(loader, cancel), nil
+	perRecord, err := pricing.NewEntries(cfg.Pricing.Providers)
+	if err != nil {
+		return nil, fmt.Errorf("cost_meter per-provider pricing: %w", err)
+	}
+	return newMiddleware(defaults, perRecord), nil
 }

-// startReloader binds the loader's mtime-poll goroutine to a context
-// derived from the proxy-lifetime context and returns its cancel func so
-// the owning middleware can stop the goroutine on teardown. Returns nil
-// when there's nothing to watch (nil context or defaults-only loader), in
-// which case the middleware's Close is a no-op.
-func startReloader(ctx context.Context, loader *pricing.Loader) context.CancelFunc {
-	if ctx == nil || !loader.WatchesFile() {
-		return nil
+// mustEmptyTable returns a valid empty table. NewTable on a nil map cannot
+// fail; the panic guard documents that invariant.
+func mustEmptyTable() *pricing.Table {
+	t, err := pricing.NewTable(nil)
+	if err != nil {
+		panic(fmt.Sprintf("cost_meter: empty pricing table must build: %v", err))
 	}
-	cctx, cancel := context.WithCancel(ctx)
-	go loader.Reload(cctx)
-	return cancel
+	return t
 }

 // decodeConfig accepts empty, null, and {} configs, returning a
--- a/proxy/internal/middleware/builtin/cost_meter/middleware.go
+++ b/proxy/internal/middleware/builtin/cost_meter/middleware.go
@@ -1,7 +1,9 @@
 // Package cost_meter implements the SlotOnResponse middleware that
 // converts token-usage metadata emitted by llm_response_parser into a
-// per-request USD cost estimate. The middleware uses the shared pricing
-// loader so operator pricing overrides apply to the chain.
+// per-request USD cost estimate. Pricing arrives from management inside
+// the middleware config: a per-provider-record table (the operator's
+// stored prices, matched via llm.resolved_provider_id) consulted first,
+// then the surface-keyed defaults table.
 package cost_meter

 import (
@@ -17,7 +19,9 @@ import (
 const ID = "cost_meter"

 // Version is the implementation version emitted via the spec merge.
-const Version = "1.0.0"
+// 1.1.0: pricing is config-delivered (defaults + per-provider-record
+// entries) instead of proxy-embedded.
+const Version = "1.1.0"

 // Skip reasons emitted under KeyCostSkipped. The set is closed; the
 // dashboard surfaces these verbatim.
@@ -42,19 +46,21 @@ var metadataKeys = []string{
 }

 // Middleware computes a per-response cost estimate from the token
-// counts emitted upstream by llm_response_parser.
+// counts emitted upstream by llm_response_parser. Both tables are
+// immutable — a pricing change arrives as a mapping push that rebuilds
+// the chain with a fresh instance.
 type Middleware struct {
-	loader *pricing.Loader
-	// cancel stops this instance's pricing-reload goroutine. Non-nil only
-	// when the loader watches an override file; Close calls it so a chain
-	// rebuild doesn't leak a poll goroutine per retired instance.
-	cancel context.CancelFunc
+	// defaults is the surface-keyed table (llm.provider x llm.model).
+	defaults *pricing.Table
+	// perRecord is keyed by provider record id (llm.resolved_provider_id)
+	// then normalized model id; entries arrive fully materialized from
+	// management. Consulted before defaults. May be nil.
+	perRecord map[string]map[string]pricing.Entry
 }

-// newMiddleware constructs a Middleware bound to the given pricing loader.
-// cancel may be nil (defaults-only loader with no reloader to stop).
-func newMiddleware(loader *pricing.Loader, cancel context.CancelFunc) *Middleware {
-	return &Middleware{loader: loader, cancel: cancel}
+// newMiddleware constructs a Middleware over the given pricing tables.
+func newMiddleware(defaults *pricing.Table, perRecord map[string]map[string]pricing.Entry) *Middleware {
+	return &Middleware{defaults: defaults, perRecord: perRecord}
 }

 // ID returns the registry identifier.
@@ -79,16 +85,9 @@ func (m *Middleware) MetadataKeys() []string {
 // response.
 func (m *Middleware) MutationsSupported() bool { return false }

-// Close stops this instance's pricing-reload goroutine, if any. Called by
-// the chain when a rebuild retires the instance, so the mtime-poll loop
-// doesn't outlive the chain it belonged to. Safe to call on a nil receiver
-// and on an instance with no reloader.
-func (m *Middleware) Close() error {
-	if m != nil && m.cancel != nil {
-		m.cancel()
-	}
-	return nil
-}
+// Close releases resources owned by the middleware. Stateless — the
+// pricing tables are plain maps owned by this instance.
+func (m *Middleware) Close() error { return nil }

 // Invoke reads provider, model, and token metadata, looks up pricing,
 // and emits either KeyCostUSDTotal or KeyCostSkipped. The decision is
@@ -144,8 +143,7 @@ func (m *Middleware) Invoke(_ context.Context, in *middleware.Input) (*middlewar
 		return out, nil
 	}

-	table := m.loader.Get()
-	costs, ok := table.Costs(provider, model, inTokens, outTokens, cachedTokens, cacheCreationTokens)
+	costs, ok := m.lookupCosts(in.Metadata, provider, model, inTokens, outTokens, cachedTokens, cacheCreationTokens)
 	if !ok {
 		out.Metadata = skip(skipUnknownModel)
 		return out, nil
@@ -164,6 +162,26 @@ func (m *Middleware) Invoke(_ context.Context, in *middleware.Input) (*middlewar
 	return out, nil
 }

+// lookupCosts resolves the price for this request and computes the cost
+// split. Resolution order:
+//
+//  1. Per-provider-record entry: the operator's stored price for the
+//     provider route that served the request, keyed by the
+//     llm.resolved_provider_id metadata llm_router stamped on the allow
+//     path. Absent metadata (e.g. no router in the chain) skips this tier.
+//  2. Surface defaults: the catalog-derived table keyed by llm.provider.
+//
+// The surface always selects the cache formula — a per-record entry for an
+// Anthropic route still bills its cache buckets additively.
+func (m *Middleware) lookupCosts(md []middleware.KV, surface, model string, inTokens, outTokens, cachedTokens, cacheCreationTokens int64) (pricing.Costs, bool) {
+	if recordID := lookupKV(md, middleware.KeyLLMResolvedProviderID); recordID != "" {
+		if entry, ok := m.perRecord[recordID][model]; ok {
+			return pricing.EntryCosts(entry, surface, inTokens, outTokens, cachedTokens, cacheCreationTokens), true
+		}
+	}
+	return m.defaults.Costs(surface, model, inTokens, outTokens, cachedTokens, cacheCreationTokens)
+}
+
 // usd renders a cost as the fixed-precision string every cost.usd_* key
 // carries, so the per-bucket values and the aggregates round identically.
 //
--- a/proxy/internal/middleware/builtin/cost_meter/middleware_test.go
+++ b/proxy/internal/middleware/builtin/cost_meter/middleware_test.go
@@ -3,39 +3,50 @@ package cost_meter
 import (
 	"context"
 	"encoding/json"
-	"os"
-	"path/filepath"
 	"testing"

 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"

+	"github.com/netbirdio/netbird/proxy/internal/llm/pricing"
 	"github.com/netbirdio/netbird/proxy/internal/middleware"
-	"github.com/netbirdio/netbird/proxy/internal/middleware/builtin"
 )

-const fixturePricing = `openai:
-  gpt-4o:
-    input_per_1k: 0.0025
-    output_per_1k: 0.01
-  gpt-4o-mini:
-    input_per_1k: 0.00015
-    output_per_1k: 0.0006
-anthropic:
-  claude-sonnet-4-5:
-    input_per_1k: 0.003
-    output_per_1k: 0.015
-`
-
-// configureBuiltin points the package-level FactoryContext at a tmp
-// directory containing the test pricing fixture. Returns the path so
-// callers can override files later if needed.
-func configureBuiltin(t *testing.T) string {
+// fixtureConfig mirrors what management's buildCostMeterConfigJSON ships:
+// a surface-keyed defaults table. Rates match the retired YAML fixture so
+// every cost assertion below is byte-identical to the pre-feature values.
+func fixtureConfig(t *testing.T) []byte {
 	t.Helper()
-	dir := t.TempDir()
-	require.NoError(t, os.WriteFile(filepath.Join(dir, "pricing.yaml"), []byte(fixturePricing), 0o600), "write pricing fixture")
-	builtin.Configure(context.Background(), dir, nil, nil, nil)
-	return dir
+	raw, err := json.Marshal(Config{Pricing: &PricingConfig{
+		Defaults: map[string]map[string]pricing.EntryJSON{
+			"openai": {
+				"gpt-4o":      {InputPer1K: 0.0025, OutputPer1K: 0.01},
+				"gpt-4o-mini": {InputPer1K: 0.00015, OutputPer1K: 0.0006},
+			},
+			"anthropic": {
+				"claude-sonnet-4-5": {InputPer1K: 0.003, OutputPer1K: 0.015},
+			},
+		},
+	}})
+	require.NoError(t, err)
+	return raw
+}
+
+// fixtureConfigWithCache adds the cache-rate fields.
+func fixtureConfigWithCache(t *testing.T) []byte {
+	t.Helper()
+	raw, err := json.Marshal(Config{Pricing: &PricingConfig{
+		Defaults: map[string]map[string]pricing.EntryJSON{
+			"openai": {
+				"gpt-4o": {InputPer1K: 0.0025, OutputPer1K: 0.01, CachedInputPer1K: 0.00125},
+			},
+			"anthropic": {
+				"claude-sonnet-4-5": {InputPer1K: 0.003, OutputPer1K: 0.015, CacheReadPer1K: 0.0003, CacheCreationPer1K: 0.00375},
+			},
+		},
+	}})
+	require.NoError(t, err)
+	return raw
 }

 func metaValue(t *testing.T, kvs []middleware.KV, key string) (string, bool) {
@@ -56,8 +67,7 @@ func buildMiddleware(t *testing.T, raw []byte) middleware.Middleware {
 }

 func TestMiddleware_StaticSurface(t *testing.T) {
-	configureBuiltin(t)
-	mw := buildMiddleware(t, nil)
+	mw := buildMiddleware(t, fixtureConfig(t))

 	assert.Equal(t, ID, mw.ID(), "ID must match the registered constant")
 	assert.Equal(t, Version, mw.Version(), "Version must match the constant")
@@ -79,8 +89,10 @@ func TestMiddleware_StaticSurface(t *testing.T) {
 	assert.Equal(t, expected, keys, "metadata key allowlist must match the spec")
 }

+// TestFactory_AcceptsEmptyAndJSONConfig: empty/null/{} configs are what an
+// old management (pre config-delivered pricing) sends — they must build a
+// working (all-skip) instance, never fail the chain.
 func TestFactory_AcceptsEmptyAndJSONConfig(t *testing.T) {
-	configureBuiltin(t)
 	cases := [][]byte{nil, {}, []byte("null"), []byte("{}"), []byte("   ")}
 	for _, raw := range cases {
 		mw, err := Factory{}.New(raw)
@@ -90,15 +102,57 @@ func TestFactory_AcceptsEmptyAndJSONConfig(t *testing.T) {
 }

 func TestFactory_RejectsMalformedConfig(t *testing.T) {
-	configureBuiltin(t)
 	mw, err := Factory{}.New([]byte("{not json"))
 	require.Error(t, err, "malformed config must surface at construction")
 	assert.Nil(t, mw, "no instance is returned on error")
 }

-func TestFactory_DefaultPricingPathLoadsFixture(t *testing.T) {
-	configureBuiltin(t)
-	mw := buildMiddleware(t, nil)
+// TestFactory_RejectsInvalidRates: a non-finite or negative rate anywhere
+// in the table fails the chain build (defense-in-depth behind management's
+// API validation) rather than silently mispricing.
+func TestFactory_RejectsInvalidRates(t *testing.T) {
+	raw, err := json.Marshal(Config{Pricing: &PricingConfig{
+		Defaults: map[string]map[string]pricing.EntryJSON{
+			"openai": {"gpt-4o": {InputPer1K: -0.0025, OutputPer1K: 0.01}},
+		},
+	}})
+	require.NoError(t, err)
+	mw, err := Factory{}.New(raw)
+	require.Error(t, err, "negative rate must fail the build")
+	assert.Nil(t, mw)
+
+	raw, err = json.Marshal(Config{Pricing: &PricingConfig{
+		Providers: map[string]map[string]pricing.EntryJSON{
+			"prov-1": {"m": {InputPer1K: 0.01, OutputPer1K: 0.01, CacheReadPer1K: -1}},
+		},
+	}})
+	require.NoError(t, err)
+	_, err = Factory{}.New(raw)
+	require.Error(t, err, "per-record tables validate too")
+}
+
+// TestFactory_NilPricingSkipsEverything is the version-skew contract: a
+// new proxy under an old management ({} config) must build, allow, and
+// skip with unknown_model — degraded but never broken.
+func TestFactory_NilPricingSkipsEverything(t *testing.T) {
+	mw := buildMiddleware(t, []byte("{}"))
+	out, err := mw.Invoke(context.Background(), &middleware.Input{
+		Metadata: []middleware.KV{
+			{Key: middleware.KeyLLMProvider, Value: "openai"},
+			{Key: middleware.KeyLLMModel, Value: "gpt-4o"},
+			{Key: middleware.KeyLLMInputTokens, Value: "1000"},
+			{Key: middleware.KeyLLMOutputTokens, Value: "1000"},
+		},
+	})
+	require.NoError(t, err)
+	assert.Equal(t, middleware.DecisionAllow, out.Decision, "cost_meter always allows")
+	value, ok := metaValue(t, out.Metadata, middleware.KeyCostSkipped)
+	require.True(t, ok, "no pricing table means every request skips")
+	assert.Equal(t, skipUnknownModel, value)
+}
+
+func TestFactory_ConfigDefaultsPriceRequests(t *testing.T) {
+	mw := buildMiddleware(t, fixtureConfig(t))

 	out, err := mw.Invoke(context.Background(), &middleware.Input{
 		Metadata: []middleware.KV{
@@ -116,34 +170,78 @@ func TestFactory_DefaultPricingPathLoadsFixture(t *testing.T) {
 	assert.Equal(t, "0.000750000", value, "0.00015 + 0.0006 per 1k tokens, 9-decimal format")
 }

-func TestFactory_PricingPathOverride(t *testing.T) {
-	dir := t.TempDir()
-	require.NoError(t, os.WriteFile(filepath.Join(dir, "custom.yaml"), []byte(fixturePricing), 0o600), "write custom pricing")
-	builtin.Configure(context.Background(), dir, nil, nil, nil)
-
-	raw, err := json.Marshal(Config{PricingPath: "custom.yaml"})
+// TestInvoke_PerRecordEntryBeatsDefaults: when llm_router resolved a
+// provider record whose operator pinned a price for the model, that price
+// wins over the surface default.
+func TestInvoke_PerRecordEntryBeatsDefaults(t *testing.T) {
+	raw, err := json.Marshal(Config{Pricing: &PricingConfig{
+		Defaults: map[string]map[string]pricing.EntryJSON{
+			"openai": {"gpt-4o": {InputPer1K: 0.0025, OutputPer1K: 0.01}},
+		},
+		Providers: map[string]map[string]pricing.EntryJSON{
+			"prov-azure": {"gpt-4o": {InputPer1K: 0.005, OutputPer1K: 0.02}},
+		},
+	}})
 	require.NoError(t, err)
-
 	mw := buildMiddleware(t, raw)
+
 	out, err := mw.Invoke(context.Background(), &middleware.Input{
 		Metadata: []middleware.KV{
 			{Key: middleware.KeyLLMProvider, Value: "openai"},
 			{Key: middleware.KeyLLMModel, Value: "gpt-4o"},
-			{Key: middleware.KeyLLMInputTokens, Value: "2000"},
+			{Key: middleware.KeyLLMResolvedProviderID, Value: "prov-azure"},
+			{Key: middleware.KeyLLMInputTokens, Value: "1000"},
 			{Key: middleware.KeyLLMOutputTokens, Value: "1000"},
 		},
 	})
 	require.NoError(t, err)
-
 	value, ok := metaValue(t, out.Metadata, middleware.KeyCostUSDTotal)
-	require.True(t, ok, "cost.usd_total must be emitted with custom pricing path")
-	assert.Equal(t, "0.015000000", value, "2*0.0025 + 1*0.01 = 0.015 with 9-decimal format")
+	require.True(t, ok)
+	assert.Equal(t, "0.025000000", value, "operator's per-record price (0.005+0.02) wins over the default (0.0025+0.01)")
 }

-func TestInvoke_ComputesCostForKnownModel(t *testing.T) {
-	configureBuiltin(t)
-	mw := buildMiddleware(t, nil)
+// TestInvoke_PerRecordMissFallsBackToDefaults: a resolved record with no
+// entry for this model (or no entries at all) falls through to the
+// surface defaults — gateway providers rely on exactly this.
+func TestInvoke_PerRecordMissFallsBackToDefaults(t *testing.T) {
+	raw, err := json.Marshal(Config{Pricing: &PricingConfig{
+		Defaults: map[string]map[string]pricing.EntryJSON{
+			"openai": {"gpt-4o": {InputPer1K: 0.0025, OutputPer1K: 0.01}},
+		},
+		Providers: map[string]map[string]pricing.EntryJSON{
+			"prov-1": {"some-other-model": {InputPer1K: 1, OutputPer1K: 1}},
+		},
+	}})
+	require.NoError(t, err)
+	mw := buildMiddleware(t, raw)

+	for name, recordID := range map[string]string{
+		"record with other models": "prov-1",
+		"record with no entries":   "prov-gateway",
+	} {
+		t.Run(name, func(t *testing.T) {
+			out, err := mw.Invoke(context.Background(), &middleware.Input{
+				Metadata: []middleware.KV{
+					{Key: middleware.KeyLLMProvider, Value: "openai"},
+					{Key: middleware.KeyLLMModel, Value: "gpt-4o"},
+					{Key: middleware.KeyLLMResolvedProviderID, Value: recordID},
+					{Key: middleware.KeyLLMInputTokens, Value: "1000"},
+					{Key: middleware.KeyLLMOutputTokens, Value: "1000"},
+				},
+			})
+			require.NoError(t, err)
+			value, ok := metaValue(t, out.Metadata, middleware.KeyCostUSDTotal)
+			require.True(t, ok, "per-record miss must fall back to the surface default, not skip")
+			assert.Equal(t, "0.012500000", value, "default rates apply")
+		})
+	}
+}
+
+// TestInvoke_NoResolvedProviderIDUsesDefaults: metadata without a
+// resolved provider id (router denied, or a chain without llm_router)
+// prices from the defaults table directly.
+func TestInvoke_NoResolvedProviderIDUsesDefaults(t *testing.T) {
+	mw := buildMiddleware(t, fixtureConfig(t))
 	out, err := mw.Invoke(context.Background(), &middleware.Input{
 		Metadata: []middleware.KV{
 			{Key: middleware.KeyLLMProvider, Value: "anthropic"},
@@ -153,17 +251,15 @@ func TestInvoke_ComputesCostForKnownModel(t *testing.T) {
 		},
 	})
 	require.NoError(t, err)
-
 	value, ok := metaValue(t, out.Metadata, middleware.KeyCostUSDTotal)
-	require.True(t, ok, "cost.usd_total must be emitted")
+	require.True(t, ok)
 	assert.Equal(t, "0.018000000", value, "0.003 + 0.015 = 0.018 with 9-decimal format")
 	_, skipped := metaValue(t, out.Metadata, middleware.KeyCostSkipped)
 	assert.False(t, skipped, "cost.skipped must not be set when cost is computed")
 }

 func TestInvoke_MissingProvider(t *testing.T) {
-	configureBuiltin(t)
-	mw := buildMiddleware(t, nil)
+	mw := buildMiddleware(t, fixtureConfig(t))

 	out, err := mw.Invoke(context.Background(), &middleware.Input{
 		Metadata: []middleware.KV{
@@ -179,8 +275,7 @@ func TestInvoke_MissingProvider(t *testing.T) {
 }

 func TestInvoke_MissingModel(t *testing.T) {
-	configureBuiltin(t)
-	mw := buildMiddleware(t, nil)
+	mw := buildMiddleware(t, fixtureConfig(t))

 	out, err := mw.Invoke(context.Background(), &middleware.Input{
 		Metadata: []middleware.KV{
@@ -196,8 +291,7 @@ func TestInvoke_MissingModel(t *testing.T) {
 }

 func TestInvoke_MissingTokens(t *testing.T) {
-	configureBuiltin(t)
-	mw := buildMiddleware(t, nil)
+	mw := buildMiddleware(t, fixtureConfig(t))

 	cases := []struct {
 		name string
@@ -240,8 +334,7 @@ func TestInvoke_MissingTokens(t *testing.T) {
 }

 func TestInvoke_UnparseableTokens(t *testing.T) {
-	configureBuiltin(t)
-	mw := buildMiddleware(t, nil)
+	mw := buildMiddleware(t, fixtureConfig(t))

 	cases := []struct {
 		name string
@@ -271,8 +364,7 @@ func TestInvoke_UnparseableTokens(t *testing.T) {
 }

 func TestInvoke_ZeroTokens(t *testing.T) {
-	configureBuiltin(t)
-	mw := buildMiddleware(t, nil)
+	mw := buildMiddleware(t, fixtureConfig(t))

 	out, err := mw.Invoke(context.Background(), &middleware.Input{
 		Metadata: []middleware.KV{
@@ -291,8 +383,7 @@ func TestInvoke_ZeroTokens(t *testing.T) {
 }

 func TestInvoke_UnknownModel(t *testing.T) {
-	configureBuiltin(t)
-	mw := buildMiddleware(t, nil)
+	mw := buildMiddleware(t, fixtureConfig(t))

 	out, err := mw.Invoke(context.Background(), &middleware.Input{
 		Metadata: []middleware.KV{
@@ -309,8 +400,7 @@ func TestInvoke_UnknownModel(t *testing.T) {
 }

 func TestInvoke_NilInput(t *testing.T) {
-	configureBuiltin(t)
-	mw := buildMiddleware(t, nil)
+	mw := buildMiddleware(t, fixtureConfig(t))

 	out, err := mw.Invoke(context.Background(), nil)
 	require.NoError(t, err)
@@ -319,36 +409,12 @@ func TestInvoke_NilInput(t *testing.T) {
 	assert.Empty(t, out.Metadata, "no metadata must be emitted on nil input")
 }

-const fixturePricingWithCache = `openai:
-  gpt-4o:
-    input_per_1k: 0.0025
-    output_per_1k: 0.01
-    cached_input_per_1k: 0.00125
-anthropic:
-  claude-sonnet-4-5:
-    input_per_1k: 0.003
-    output_per_1k: 0.015
-    cache_read_per_1k: 0.0003
-    cache_creation_per_1k: 0.00375
-`
-
-// configureBuiltinWithCacheRates points the package-level
-// FactoryContext at a tmp directory containing pricing entries that
-// include the cache rate fields.
-func configureBuiltinWithCacheRates(t *testing.T) {
-	t.Helper()
-	dir := t.TempDir()
-	require.NoError(t, os.WriteFile(filepath.Join(dir, "pricing.yaml"), []byte(fixturePricingWithCache), 0o600), "write cache-aware pricing fixture")
-	builtin.Configure(context.Background(), dir, nil, nil, nil)
-}
-
 // TestInvoke_OpenAICachedSubsetDiscount proves the OpenAI shape end
 // to end through the middleware: cached_input_tokens is treated as a
 // SUBSET of input_tokens and discounted at the configured rate, not
 // added on top.
 func TestInvoke_OpenAICachedSubsetDiscount(t *testing.T) {
-	configureBuiltinWithCacheRates(t)
-	mw := buildMiddleware(t, nil)
+	mw := buildMiddleware(t, fixtureConfigWithCache(t))

 	out, err := mw.Invoke(context.Background(), &middleware.Input{
 		Metadata: []middleware.KV{
@@ -390,8 +456,7 @@ func TestInvoke_OpenAICachedSubsetDiscount(t *testing.T) {
 // shape: cache_read and cache_creation are additive to input_tokens
 // and each carries its own rate.
 func TestInvoke_AnthropicCacheBucketsAdditive(t *testing.T) {
-	configureBuiltinWithCacheRates(t)
-	mw := buildMiddleware(t, nil)
+	mw := buildMiddleware(t, fixtureConfigWithCache(t))

 	out, err := mw.Invoke(context.Background(), &middleware.Input{
 		Metadata: []middleware.KV{
@@ -429,8 +494,37 @@ func TestInvoke_AnthropicCacheBucketsAdditive(t *testing.T) {
 		"output bucket bills 200 tokens at 0.015/1k")
 }

+// TestInvoke_PerRecordEntryUsesSurfaceFormula: a per-record entry for an
+// anthropic-surface request must bill its cache buckets additively — the
+// formula follows llm.provider, not which table the entry came from.
+func TestInvoke_PerRecordEntryUsesSurfaceFormula(t *testing.T) {
+	raw, err := json.Marshal(Config{Pricing: &PricingConfig{
+		Providers: map[string]map[string]pricing.EntryJSON{
+			"prov-ant": {"claude-sonnet-4-5": {InputPer1K: 0.003, OutputPer1K: 0.015, CacheReadPer1K: 0.0003, CacheCreationPer1K: 0.00375}},
+		},
+	}})
+	require.NoError(t, err)
+	mw := buildMiddleware(t, raw)
+
+	out, err := mw.Invoke(context.Background(), &middleware.Input{
+		Metadata: []middleware.KV{
+			{Key: middleware.KeyLLMProvider, Value: "anthropic"},
+			{Key: middleware.KeyLLMModel, Value: "claude-sonnet-4-5"},
+			{Key: middleware.KeyLLMResolvedProviderID, Value: "prov-ant"},
+			{Key: middleware.KeyLLMInputTokens, Value: "256"},
+			{Key: middleware.KeyLLMOutputTokens, Value: "200"},
+			{Key: middleware.KeyLLMCachedInputTokens, Value: "768"},
+			{Key: middleware.KeyLLMCacheCreationTokens, Value: "512"},
+		},
+	})
+	require.NoError(t, err)
+	value, ok := metaValue(t, out.Metadata, middleware.KeyCostUSDTotal)
+	require.True(t, ok)
+	assert.Equal(t, "0.005918400", value, "identical math to the defaults-table entry with the same rates")
+}
+
 // assertBucket asserts one per-bucket cost key carries the expected
-// 6-decimal value.
+// 9-decimal value.
 func assertBucket(t *testing.T, md []middleware.KV, key, want, msg string) {
 	t.Helper()
 	got, ok := metaValue(t, md, key)
@@ -439,13 +533,10 @@ func assertBucket(t *testing.T, md []middleware.KV, key, want, msg string) {
 }

 // TestInvoke_CachedTokensAbsentFallsBackToBaseFormula covers the
-// "operator hasn't opted in" path: with no cached metadata keys
-// emitted, the meter must produce exactly the same cost as before
-// the feature landed. Critical so operators with the new binary but
-// no YAML changes see no behavioural drift on OpenAI requests.
+// no-cache-metadata path: with no cached keys emitted, the meter must
+// produce exactly the input+output cost.
 func TestInvoke_CachedTokensAbsentFallsBackToBaseFormula(t *testing.T) {
-	configureBuiltinWithCacheRates(t)
-	mw := buildMiddleware(t, nil)
+	mw := buildMiddleware(t, fixtureConfigWithCache(t))

 	out, err := mw.Invoke(context.Background(), &middleware.Input{
 		Metadata: []middleware.KV{
@@ -460,7 +551,7 @@ func TestInvoke_CachedTokensAbsentFallsBackToBaseFormula(t *testing.T) {
 	value, ok := metaValue(t, out.Metadata, middleware.KeyCostUSDTotal)
 	require.True(t, ok)
 	// 1000 input * 0.0025 + 500 output * 0.01 = 0.0025 + 0.005 = 0.0075
-	assert.Equal(t, "0.007500000", value, "no cached metadata = same cost as before the feature landed")
+	assert.Equal(t, "0.007500000", value, "no cached metadata = plain input+output cost")
 }

 // TestInvoke_UnparseableCachedTokensSkippedSilently proves the
@@ -469,8 +560,7 @@ func TestInvoke_CachedTokensAbsentFallsBackToBaseFormula(t *testing.T) {
 // regular formula. Cache buckets are a refinement, never a reason to
 // abort cost computation.
 func TestInvoke_UnparseableCachedTokensSkippedSilently(t *testing.T) {
-	configureBuiltinWithCacheRates(t)
-	mw := buildMiddleware(t, nil)
+	mw := buildMiddleware(t, fixtureConfigWithCache(t))

 	out, err := mw.Invoke(context.Background(), &middleware.Input{
 		Metadata: []middleware.KV{
@@ -487,22 +577,10 @@ func TestInvoke_UnparseableCachedTokensSkippedSilently(t *testing.T) {
 	assert.Equal(t, "0.007500000", value, "same as the no-cached-metadata path")
 }

-// TestMiddleware_CloseCancelsReloader proves Close stops the per-instance
-// pricing-reload goroutine: a chain rebuild retires the old instance and
-// calls Close, which must invoke the cancel func startReloader handed it so
-// the mtime-poll loop doesn't outlive the chain.
-func TestMiddleware_CloseCancelsReloader(t *testing.T) {
-	ctx, cancel := context.WithCancel(context.Background())
-	m := newMiddleware(nil, cancel)
-
-	require.NoError(t, m.Close(), "Close must not error")
-	require.Error(t, ctx.Err(), "Close must cancel the reloader context so the poll goroutine exits")
-}
-
-// TestMiddleware_CloseNilSafe confirms Close is a no-op (no panic) for an
-// instance with no reloader and for a nil receiver.
+// TestMiddleware_CloseNilSafe confirms Close is a no-op (no panic) even
+// for a nil receiver.
 func TestMiddleware_CloseNilSafe(t *testing.T) {
-	require.NoError(t, newMiddleware(nil, nil).Close(), "no-reloader Close must be a no-op")
+	require.NoError(t, newMiddleware(nil, nil).Close(), "Close must be a no-op")
 	var m *Middleware
 	require.NoError(t, m.Close(), "nil-receiver Close must be safe")
 }
--- a/proxy/internal/middleware/builtin/llm_request_parser/bedrock_test.go
+++ b/proxy/internal/middleware/builtin/llm_request_parser/bedrock_test.go
@@ -6,23 +6,6 @@ import (
 	"github.com/stretchr/testify/require"
 )

-func TestNormalizeBedrockModel(t *testing.T) {
-	cases := map[string]string{
-		"eu.anthropic.claude-sonnet-4-5-20250929-v1:0": "anthropic.claude-sonnet-4-5",
-		"us.anthropic.claude-opus-4-8-20250101-v1:0":   "anthropic.claude-opus-4-8",
-		"apac.anthropic.claude-haiku-4-5-v1:0":         "anthropic.claude-haiku-4-5",
-		"anthropic.claude-sonnet-4-5-20250929-v1:0":    "anthropic.claude-sonnet-4-5",
-		"meta.llama3-3-70b-instruct-v1:0":              "meta.llama3-3-70b-instruct",
-		"amazon.nova-pro-v1:0":                         "amazon.nova-pro",
-		"amazon.nova-2-lite-v1:0":                      "amazon.nova-2-lite",
-		// Inference-profile ARN — model id lives in the last path segment.
-		"arn:aws:bedrock:eu-central-1:123456789012:inference-profile/eu.anthropic.claude-sonnet-4-5-20250929-v1:0": "anthropic.claude-sonnet-4-5",
-	}
-	for in, want := range cases {
-		require.Equal(t, want, normalizeBedrockModel(in), "normalize %q", in)
-	}
-}
-
 func TestParseBedrockPath(t *testing.T) {
 	tests := []struct {
 		path   string
--- a/proxy/internal/middleware/builtin/llm_request_parser/middleware.go
+++ b/proxy/internal/middleware/builtin/llm_request_parser/middleware.go
@@ -8,7 +8,6 @@ package llm_request_parser
 import (
 	"context"
 	"net/url"
-	"regexp"
 	"strconv"
 	"strings"
 	"unicode/utf8"
@@ -253,9 +252,7 @@ func parseVertexPath(reqPath string) (vertexRequest, bool) {
 	if c := strings.LastIndex(rest, ":"); c >= 0 {
 		model, action = rest[:c], rest[c+1:]
 	}
-	if at := strings.Index(model, "@"); at >= 0 {
-		model = model[:at]
-	}
+	model = llm.NormalizeVertexModel(model)
 	if model == "" {
 		return vertexRequest{}, false
 	}
@@ -343,14 +340,6 @@ func trimBedrockNamespace(reqPath string) string {
 	return reqPath
 }

-// bedrockRegionPrefixes are the cross-region inference-profile prefixes that
-// front a Bedrock model id (e.g. "eu.anthropic.claude-...").
-var bedrockRegionPrefixes = []string{"us.", "eu.", "apac.", "global."}
-
-// bedrockVersionSuffix matches the trailing "-vN[:N]" or "-YYYYMMDD-vN[:N]"
-// version/throughput suffix of a Bedrock model id.
-var bedrockVersionSuffix = regexp.MustCompile(`-(\d{8}-)?v\d+(:\d+)?$`)
-
 // parseBedrockPath extracts the model and streaming/converse flags from an AWS
 // Bedrock runtime model endpoint:
 //
@@ -375,7 +364,7 @@ func parseBedrockPath(reqPath string) (bedrockRequest, bool) {
 	if decoded, err := url.PathUnescape(rawModel); err == nil {
 		rawModel = decoded
 	}
-	model := normalizeBedrockModel(rawModel)
+	model := llm.NormalizeBedrockModel(rawModel)
 	if model == "" {
 		return bedrockRequest{}, false
 	}
@@ -389,30 +378,6 @@ func parseBedrockPath(reqPath string) (bedrockRequest, bool) {
 	}
 }

-// normalizeBedrockModel strips an ARN wrapper, a cross-region inference-profile
-// prefix, and the version/throughput suffix from a Bedrock model id so it
-// matches the catalog/pricing key, e.g.
-// "eu.anthropic.claude-sonnet-4-5-20250929-v1:0" -> "anthropic.claude-sonnet-4-5"
-// and "arn:aws:bedrock:eu-central-1:123:inference-profile/eu.anthropic.claude-sonnet-4-5-20250929-v1:0"
-// -> "anthropic.claude-sonnet-4-5".
-func normalizeBedrockModel(modelID string) string {
-	m := modelID
-	// A full ARN (inference-profile / provisioned-throughput / foundation-model)
-	// carries the model id in its last path segment.
-	if strings.HasPrefix(m, "arn:") {
-		if i := strings.LastIndex(m, "/"); i >= 0 {
-			m = m[i+1:]
-		}
-	}
-	for _, p := range bedrockRegionPrefixes {
-		if strings.HasPrefix(m, p) {
-			m = m[len(p):]
-			break
-		}
-	}
-	return bedrockVersionSuffix.ReplaceAllString(m, "")
-}
-
 // invokeBedrock emits the model/provider/session/prompt for an AWS Bedrock
 // request. Bedrock is metered under the dedicated "bedrock" parser, which reads
 // both the InvokeModel and Converse response shapes.
--- a/proxy/internal/proxy/agent_network_chain_realstack_test.go
+++ b/proxy/internal/proxy/agent_network_chain_realstack_test.go
@@ -18,10 +18,10 @@ import (
 	"google.golang.org/grpc/credentials/insecure"
 	"google.golang.org/grpc/test/bufconn"

-	rpservice "github.com/netbirdio/netbird/management/internals/modules/reverseproxy/service"
-	mgmtgrpc "github.com/netbirdio/netbird/management/internals/shared/grpc"
 	"github.com/netbirdio/netbird/management/internals/modules/agentnetwork"
 	agentNetworkTypes "github.com/netbirdio/netbird/management/internals/modules/agentnetwork/types"
+	rpservice "github.com/netbirdio/netbird/management/internals/modules/reverseproxy/service"
+	mgmtgrpc "github.com/netbirdio/netbird/management/internals/shared/grpc"
 	"github.com/netbirdio/netbird/management/server/store"
 	"github.com/netbirdio/netbird/proxy/internal/middleware"
 	"github.com/netbirdio/netbird/proxy/internal/middleware/bodytap"
@@ -134,14 +134,18 @@ func TestReverseProxy_AgentNetworkRequest_FullChain(t *testing.T) {
 		RedactPii:              true,
 	}))
 	require.NoError(t, st.SaveAgentNetworkProvider(ctx, &agentNetworkTypes.Provider{
-		ID:                providerID,
-		AccountID:         testAccountID,
-		ProviderID:        "openai_api",
-		Name:              "openai-fullchain-test",
-		UpstreamURL:       upstream.URL, // router rewrites to this
-		APIKey:            "sk-test",
-		Enabled:           true,
-		Models:            []agentNetworkTypes.ProviderModel{{ID: "gpt-5.4"}},
+		ID:          providerID,
+		AccountID:   testAccountID,
+		ProviderID:  "openai_api",
+		Name:        "openai-fullchain-test",
+		UpstreamURL: upstream.URL, // router rewrites to this
+		APIKey:      "sk-test",
+		Enabled:     true,
+		// Operator-pinned prices deliberately differ from the catalog's
+		// gpt-5.4 rates (0.0025/0.015) so the cost assertion below proves
+		// the per-provider-record price — not the default table — billed
+		// this request: stored price → synth → wire → cost_meter.
+		Models:            []agentNetworkTypes.ProviderModel{{ID: "gpt-5.4", InputPer1k: 0.004, OutputPer1k: 0.02}},
 		SessionPrivateKey: "priv",
 		SessionPublicKey:  "pub",
 	}))
@@ -176,7 +180,7 @@ func TestReverseProxy_AgentNetworkRequest_FullChain(t *testing.T) {

 	// ---- 5. Wire the middleware framework — same registry the proxy uses
 	// in production, configured with our bufconn-backed management client.
-	mwbuiltin.Configure(ctx, t.TempDir(), nil, testLogger, mgmtClient)
+	mwbuiltin.Configure(ctx, nil, testLogger, mgmtClient)
 	registry := mwbuiltin.DefaultRegistry()
 	mwMetrics, err := middleware.NewMetrics(nil)
 	require.NoError(t, err)
@@ -283,13 +287,23 @@ func TestReverseProxy_AgentNetworkRequest_FullChain(t *testing.T) {
 			if r.DimensionKind == agentNetworkTypes.DimensionGroup &&
 				r.DimensionID == adminGroupID &&
 				r.WindowSeconds == 60 &&
-				r.TokensInput+r.TokensOutput > 0 {
+				r.TokensInput+r.TokensOutput > 0 &&
+				r.CostUSD > 0 {
 				return true
 			}
 		}
 		return false
 	}, 5*time.Second, 50*time.Millisecond,
-		"Admins group consumption row must increment via the response leg — if this fails the proxy's respInput dropped UserGroups again or the parser/recorder wiring is broken")
+		"Admins group consumption row must increment via the response leg WITH a non-zero cost — a zero cost means the operator's stored price never reached cost_meter (synth → wire → per-record lookup broken)")
+
+	// 8a-cost. Exact cost from the OPERATOR's stored price, not the catalog
+	// default: 12 prompt tokens × 0.004/1k + 40 completion tokens × 0.02/1k
+	// = 0.000048 + 0.0008 = 0.000848. With catalog rates it would be 0.00063
+	// — this assertion distinguishes the two, closing the loop on the whole
+	// dynamic-pricing feature (dashboard save → synth → gRPC wire →
+	// llm.resolved_provider_id lookup → billing).
+	assert.Equal(t, "0.000848000", cd.GetMetadata()["cost.usd_total"],
+		"cost must be computed from the provider record's operator-pinned price")

 	// 8b. Both the captured prompt and the captured completion are
 	// redacted — proves the synth threads redact_pii=true into BOTH parser
--- a/proxy/server.go
+++ b/proxy/server.go
@@ -246,10 +246,6 @@ type Server struct {
 	// in processMappings before the receive loop reconnects to resync.
 	// Zero uses defaultMappingBatchWatchdog.
 	MappingBatchWatchdog time.Duration
-	// MiddlewareDataDir is the base directory the middleware system uses to
-	// resolve file-backed configuration (e.g. the cost_meter pricing table).
-	// Empty means any middleware that requires a file fails at configure time.
-	MiddlewareDataDir string
 	// MiddlewareCaptureBudgetBytes overrides the proxy-wide in-flight capture
 	// budget passed to middleware.NewManager. Zero or negative values fall
 	// back to defaultMiddlewareCaptureBudgetBytes (256 MiB).
@@ -2093,7 +2089,7 @@ func (s *Server) initMiddlewareManager(ctx context.Context) error {
 		return fmt.Errorf("middleware manager requires metrics bundle")
 	}
 	otelMeter := s.meter.Meter()
-	mwbuiltin.Configure(ctx, s.MiddlewareDataDir, otelMeter, s.Logger, s.mgmtClient)
+	mwbuiltin.Configure(ctx, otelMeter, s.Logger, s.mgmtClient)

 	mwMetrics, err := middleware.NewMetrics(otelMeter)
 	if err != nil {