From 2a72c9505c8a9d6af1363a7d7e67bca9bfd732fb Mon Sep 17 00:00:00 2001
From: eeshsaxena <eeshsaxena@gmail.com>
Date: Thu, 23 Jul 2026 17:22:23 +0530
Subject: [PATCH] fix(cost): tier Anthropic cache reads by the whole prompt
 size

getThresholdValueFunction returns 0 for cachedInputCost on the anthropic
branch, so getPricingTier always resolves to tier 0 and a cache read is
priced at the base input rate no matter how large the prompt is. inputCost
and outputCost on the same request already use the prompt total.

For claude-sonnet-4 a 250K-token prompt with a 240K cache read is billed
$0.072 instead of $0.144 - half rate - because the >200K tier doubles the
input rate the cache multiplier applies to.

vertex, google-ai-studio and xai all include cachedInputCost in their
threshold cases, and the Gemini over-threshold test asserts the higher tier
for its cache read, so anthropic looks like an oversight rather than intent.

Updates the existing over-threshold assertion, which had encoded the old
behaviour, and adds a mostly-cached long prompt as a regression test.
---
 .../cost/modelCostFromRegistry.test.ts        | 31 ++++++++++++++++++-
 packages/cost/models/calculate-cost.ts        | 10 ++++--
 2 files changed, 37 insertions(+), 4 deletions(-)

diff --git a/packages/__tests__/cost/modelCostFromRegistry.test.ts b/packages/__tests__/cost/modelCostFromRegistry.test.ts
index 73aa4f6602..dfaac9b114 100644
--- a/packages/__tests__/cost/modelCostFromRegistry.test.ts
+++ b/packages/__tests__/cost/modelCostFromRegistry.test.ts
@@ -317,11 +317,40 @@ describe("modelCostBreakdownFromRegistry", () => {
         // Higher tier: $6/M input, $22.50/M output (multipliers inherited from base tier)
         expect(breakdown.inputCost).toBe(250000 * 0.000006);
         expect(breakdown.outputCost).toBe(50000 * 0.0000225);
-        expect(breakdown.cachedInputCost).toBe(10000 * 0.000003 * 0.1);
+        // The cache read belongs to a prompt that is over the threshold, so it
+        // bills at the same higher tier as the input - matching how the Gemini
+        // case below prices its cache read.
+        expect(breakdown.cachedInputCost).toBe(10000 * 0.000006 * 0.1);
+        // Cache writes still use the base tier: calculateModelCostBreakdown
+        // prices those from basePricing directly, for every provider.
         expect(breakdown.cacheWrite5mCost).toBe(5000 * 0.000003 * 1.25);
       }
     });
 
+    it("should tier a Claude cache read by the whole prompt size", () => {
+      // Regression: the cache read was priced from tier 0 regardless of prompt
+      // size, because the anthropic threshold function returned 0 for
+      // cachedInputCost. A mostly-cached long prompt was billed at half rate.
+      const modelUsage: ModelUsage = {
+        input: 10000,
+        output: 100,
+        cacheDetails: {
+          cachedInput: 240000, // prompt totals 250K, over the 200K threshold
+        },
+      };
+
+      const breakdown = modelCostBreakdownFromRegistry({
+        modelUsage,
+        providerModelId: "claude-sonnet-4-20250514",
+        provider: "anthropic" as ModelProviderName,
+      });
+
+      expect(breakdown).not.toBeNull();
+      if (breakdown) {
+        expect(breakdown.cachedInputCost).toBe(240000 * 0.000006 * 0.1);
+      }
+    });
+
     it("should use base tier pricing for Gemini 3 Pro Preview under 200K tokens", () => {
       const modelUsage: ModelUsage = {
         input: 150000, // 150K tokens - under threshold
diff --git a/packages/cost/models/calculate-cost.ts b/packages/cost/models/calculate-cost.ts
index 860396a924..4b102e24a2 100644
--- a/packages/cost/models/calculate-cost.ts
+++ b/packages/cost/models/calculate-cost.ts
@@ -142,9 +142,13 @@ function getThresholdValueFunction(provider: ModelProviderName): (usage: ModelUs
         switch (field) {
           case "inputCost":
           case "outputCost":
-            return usage.input + 
-              (usage.cacheDetails?.cachedInput ?? 0) + 
-              (usage.cacheDetails?.write5m ?? 0) + 
+          // Anthropic's long-context tier is chosen by the size of the whole
+          // prompt, so a cache read on a >threshold request is billed at the
+          // same higher tier as the input it belongs to.
+          case "cachedInputCost":
+            return usage.input +
+              (usage.cacheDetails?.cachedInput ?? 0) +
+              (usage.cacheDetails?.write5m ?? 0) +
               (usage.cacheDetails?.write1h ?? 0);
           default:
             return 0;