local
54 lines
2.3 KiB
JavaScript
54 lines
2.3 KiB
JavaScript
// Gemini per-1M-token pricing in USD. Used by the relay to compute
|
|
// per-call API costs as audit-log entries are written. Operator can
|
|
// update this table by editing the file and redeploying when Google
|
|
// changes published rates — preferable to a config field because
|
|
// (a) prices are stable for months at a time, (b) hardcoding keeps
|
|
// the audit log self-contained without a config-snapshot copy at
|
|
// write time.
|
|
//
|
|
// Rates as of mid-2026. ALWAYS verify against the current Google AI
|
|
// Studio pricing page before relying on these for billing-grade
|
|
// margin math — Google has been known to adjust prices ~quarterly.
|
|
|
|
export const GEMINI_PRICING = {
|
|
// Pro family — best for analysis.
|
|
"gemini-3.1-pro-preview": { input: 5.0, output: 25.0, thinking: 25.0 },
|
|
"gemini-3-pro-preview": { input: 5.0, output: 25.0, thinking: 25.0 },
|
|
|
|
// Flash family — best speed/cost for transcription, common for
|
|
// analysis when sub-Pro quality is acceptable.
|
|
"gemini-3-flash-preview": { input: 0.3, output: 2.5, thinking: 2.5 },
|
|
"gemini-2.5-flash": { input: 0.3, output: 2.5, thinking: 2.5 },
|
|
"gemini-2.0-flash": { input: 0.1, output: 0.4, thinking: 0.4 },
|
|
|
|
// Fallback used when an unknown model id appears (e.g. operator
|
|
// typed a custom model name in setBackendRouting). Conservative —
|
|
// priced like Flash so cost estimates skew low rather than 0.
|
|
default: { input: 0.3, output: 2.5, thinking: 2.5 },
|
|
};
|
|
|
|
// Compute cost for a Gemini API call given its model + usageMetadata
|
|
// (the shape @google/genai returns: promptTokenCount,
|
|
// candidatesTokenCount, thoughtsTokenCount). Returns:
|
|
// { input_tokens, output_tokens, thinking_tokens, cost_usd }
|
|
export function calcGeminiCost(model, usage) {
|
|
const rates = GEMINI_PRICING[model] || GEMINI_PRICING.default;
|
|
const inputTokens = usage?.promptTokenCount || 0;
|
|
const outputTokens = usage?.candidatesTokenCount || 0;
|
|
const thinkingTokens = usage?.thoughtsTokenCount || 0;
|
|
const costUSD =
|
|
(inputTokens / 1_000_000) * rates.input +
|
|
(outputTokens / 1_000_000) * rates.output +
|
|
(thinkingTokens / 1_000_000) * (rates.thinking ?? rates.output);
|
|
return {
|
|
input_tokens: inputTokens,
|
|
output_tokens: outputTokens,
|
|
thinking_tokens: thinkingTokens,
|
|
cost_usd: costUSD,
|
|
};
|
|
}
|
|
|
|
export function listKnownModels() {
|
|
return Object.keys(GEMINI_PRICING).filter((k) => k !== "default");
|
|
}
|