Skip to content

fix: correct OpenAI cached input cost and fix Anthropic token double-counting #13315

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
dajiaohuang wants to merge 2 commits into
base: main
Choose a base branch
from
Open

fix: correct OpenAI cached input cost and fix Anthropic token double-counting #13315

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
eb9b910
fix: correct OpenAI cached input cost calculation and fix Anthropic d…
dajiaohuang Sep 25, 2026
0a25096
fix: correct cache usage cost fixtures
dajiaohuang Oct 2, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
118 changes: 58 additions & 60 deletions core/llm/utils/calculateRequestCost.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -86,56 +86,52 @@ function calculateAnthropicCost(
}

if (!modelPricing) {
return null; // Unknown model
return null;
}

// Calculate costs
const inputCost = (usage.promptTokens / 1_000_000) * modelPricing.input;
const cachedTokens = usage.promptTokensDetails?.cachedTokens ?? 0;
const cacheWriteTokens = usage.promptTokensDetails?.cacheWriteTokens ?? 0;
const uncachedInputTokens = Math.max(
0,
usage.promptTokens - cachedTokens - cacheWriteTokens,
);

const uncachedInputCost =
(uncachedInputTokens / 1_000_000) * modelPricing.input;
const cacheWriteCost =
(cacheWriteTokens / 1_000_000) * modelPricing.cacheWrite;
const cacheReadCost = (cachedTokens / 1_000_000) * modelPricing.cacheRead;
const outputCost = (usage.completionTokens / 1_000_000) * modelPricing.output;

// Build breakdown components
const totalCost =
uncachedInputCost + cacheWriteCost + cacheReadCost + outputCost;

const breakdownParts: string[] = [];

// Input tokens breakdown
if (usage.promptTokens > 0) {
if (uncachedInputTokens > 0) {
breakdownParts.push(
`Input: ${usage.promptTokens.toLocaleString()} tokens × $${modelPricing.input}/MTok = $${inputCost.toFixed(6)}`,
`Input: ${uncachedInputTokens.toLocaleString()} tokens × $${modelPricing.input}/MTok = $${uncachedInputCost.toFixed(6)}`,
);
}

// Output tokens breakdown
if (usage.completionTokens > 0) {
if (cacheWriteTokens > 0) {
breakdownParts.push(
`Output: ${usage.completionTokens.toLocaleString()} tokens × $${modelPricing.output}/MTok = $${outputCost.toFixed(6)}`,
`Cache Write: ${cacheWriteTokens.toLocaleString()} tokens × $${modelPricing.cacheWrite}/MTok = $${cacheWriteCost.toFixed(6)}`,
);
}

// Handle prompt caching costs if available
let cacheCost = 0;
if (usage.promptTokensDetails) {
const { cachedTokens, cacheWriteTokens } = usage.promptTokensDetails;

if (cacheWriteTokens && cacheWriteTokens > 0) {
const cacheWriteCost =
(cacheWriteTokens / 1_000_000) * modelPricing.cacheWrite;
cacheCost += cacheWriteCost;
breakdownParts.push(
`Cache Write: ${cacheWriteTokens.toLocaleString()} tokens × $${modelPricing.cacheWrite}/MTok = $${cacheWriteCost.toFixed(6)}`,
);
}

if (cachedTokens && cachedTokens > 0) {
const cacheReadCost = (cachedTokens / 1_000_000) * modelPricing.cacheRead;
cacheCost += cacheReadCost;
breakdownParts.push(
`Cache Read: ${cachedTokens.toLocaleString()} tokens × $${modelPricing.cacheRead}/MTok = $${cacheReadCost.toFixed(6)}`,
);
}
if (cachedTokens > 0) {
breakdownParts.push(
`Cache Read: ${cachedTokens.toLocaleString()} tokens × $${modelPricing.cacheRead}/MTok = $${cacheReadCost.toFixed(6)}`,
);
}

const totalCost = inputCost + outputCost + cacheCost;
if (usage.completionTokens > 0) {
breakdownParts.push(
`Output: ${usage.completionTokens.toLocaleString()} tokens × $${modelPricing.output}/MTok = $${outputCost.toFixed(6)}`,
);
}

// Build final breakdown string
let breakdown = `Model: ${model}\n`;
breakdown += breakdownParts.join("\n");
if (breakdownParts.length > 1) {
Expand All @@ -152,28 +148,21 @@ function calculateOpenAICost(
model: string,
usage: Usage,
): CostBreakdown | null {
// Normalize model name
const normalizedModel = model.toLowerCase();

// Define pricing per million tokens (MTok) by model family prefix
const pricing: Record<string, { input: number; output: number }> = {
// GPT-4o models (most specific first)
"gpt-4o-mini": { input: 0.15, output: 0.6 },
"gpt-4o": { input: 2.5, output: 10 },

// GPT-4 Turbo models
"gpt-4-turbo": { input: 10, output: 30 },

// GPT-3.5 Turbo models (most specific first)
"gpt-3.5-turbo-0125": { input: 0.5, output: 1.5 },
"gpt-3.5-turbo-1106": { input: 1, output: 2 },
"gpt-3.5-turbo": { input: 1.5, output: 2 },

// Base GPT-4 (fallback for other gpt-4 variants)
"gpt-4": { input: 30, output: 60 },
const pricing: Record<
string,
{ input: number; output: number; cachedInput: number }
> = {
"gpt-4o-mini": { input: 0.15, output: 0.6, cachedInput: 0.075 },
"gpt-4o": { input: 2.5, output: 10, cachedInput: 1.25 },
"gpt-4-turbo": { input: 10, output: 30, cachedInput: 5 },
"gpt-3.5-turbo-0125": { input: 0.5, output: 1.5, cachedInput: 0.25 },
"gpt-3.5-turbo-1106": { input: 1, output: 2, cachedInput: 0.5 },
"gpt-3.5-turbo": { input: 1.5, output: 2, cachedInput: 0.75 },
"gpt-4": { input: 30, output: 60, cachedInput: 15 },
};

// Sort keys by length (longest first) to match most specific patterns first
const sortedKeys = Object.keys(pricing).sort((a, b) => b.length - a.length);

let modelPricing = null;
Expand All @@ -185,19 +174,31 @@ function calculateOpenAICost(
}

if (!modelPricing) {
return null; // Unknown model
return null;
}

// Calculate costs
const inputCost = (usage.promptTokens / 1_000_000) * modelPricing.input;
const cachedTokens = usage.promptTokensDetails?.cachedTokens ?? 0;
const uncachedInputTokens = Math.max(0, usage.promptTokens - cachedTokens);

const uncachedInputCost =
(uncachedInputTokens / 1_000_000) * modelPricing.input;
const cachedInputCost = (cachedTokens / 1_000_000) * modelPricing.cachedInput;
const outputCost = (usage.completionTokens / 1_000_000) * modelPricing.output;

// Build breakdown components
const inputCost = uncachedInputCost + cachedInputCost;
const totalCost = inputCost + outputCost;

const breakdownParts: string[] = [];

if (usage.promptTokens > 0) {
if (uncachedInputTokens > 0) {
breakdownParts.push(
`Input: ${usage.promptTokens.toLocaleString()} tokens × $${modelPricing.input}/MTok = $${inputCost.toFixed(6)}`,
`Input: ${uncachedInputTokens.toLocaleString()} tokens × $${modelPricing.input}/MTok = $${uncachedInputCost.toFixed(6)}`,
);
}

if (cachedTokens > 0) {
breakdownParts.push(
`Cached Input: ${cachedTokens.toLocaleString()} tokens × $${modelPricing.cachedInput}/MTok = $${cachedInputCost.toFixed(6)}`,
);
}

Expand All @@ -207,9 +208,6 @@ function calculateOpenAICost(
);
}

const totalCost = inputCost + outputCost;

// Build final breakdown string
let breakdown = `Model: ${model}\n`;
breakdown += breakdownParts.join("\n");
if (breakdownParts.length > 1) {
Expand Down
4 changes: 2 additions & 2 deletions core/llm/utils/calculateRequestCost.vitest.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,7 +29,7 @@
{
provider: "anthropic",
model: "claude-3-5-sonnet-20241022",
promptTokens: 1000,
promptTokens: 3000,
completionTokens: 500,
cachedTokens: 2000,
expectedCost: 0.0111,
Expand All @@ -38,7 +38,7 @@
{
provider: "anthropic",
model: "claude-3-5-sonnet-20241022",
promptTokens: 1000,
promptTokens: 1300,
completionTokens: 500,
cacheWriteTokens: 300,
expectedCost: 0.011625,
Expand Down Expand Up @@ -248,7 +248,7 @@
expect(result).toBeNull();
} else {
expect(result).not.toBeNull();
expect(result!.cost).toBeCloseTo(expectedCost, 6);

Check failure on line 251 in core/llm/utils/calculateRequestCost.vitest.ts

View workflow job for this annotation

GitHub Actions / core-checks

llm/utils/calculateRequestCost.vitest.ts > calculateRequestCost > Claude Opus 4.5 with cache writes 

AssertionError: expected 0.017875000000000002 to be close to 0.019375, received difference is 0.0014999999999999979, but expected 5e-7
 ❯ llm/utils/calculateRequestCost.vitest.ts:251:32

Check failure on line 251 in core/llm/utils/calculateRequestCost.vitest.ts

View workflow job for this annotation

GitHub Actions / core-checks

llm/utils/calculateRequestCost.vitest.ts > calculateRequestCost > Claude Opus 4.5 with cache reads 

AssertionError: expected 0.013500000000000002 to be close to 0.0185, received difference is 0.0049999999999999975, but expected 5e-7
 ❯ llm/utils/calculateRequestCost.vitest.ts:251:32
}
});
},
Expand Down
Loading