支持导入定价文档并在统计页估算费用。
把用量与单价对齐，方便对照各模型实际开销。 Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
Cursor
@@ -0,0 +1,299 @@
|
||||
import { describe, expect, it, beforeAll } from "vitest";
|
||||
import { mkdtempSync } from "node:fs";
|
||||
import { tmpdir } from "node:os";
|
||||
import path from "node:path";
|
||||
import {
|
||||
parseLegacyAliases,
|
||||
parsePeakSchedule,
|
||||
parsePricingDocument,
|
||||
} from "../src/stats/pricing-parse.js";
|
||||
import { buildCostReport, costForRecord, isPeakAt, splitPromptTokens } from "../src/stats/pricing-cost.js";
|
||||
import type { ResolvedRate } from "../src/stats/pricing.js";
|
||||
import type { TokenUsageRecord } from "../src/stats/token-store.js";
|
||||
|
||||
const DEEPSEEK_HTML = `<!doctype html>
|
||||
<html><head><title>Models & Pricing | DeepSeek API Docs</title></head>
|
||||
<body>
|
||||
<p>The prices listed below are in units of per 1M tokens.</p>
|
||||
<table style="text-align:center">
|
||||
<tr><td colspan="3">MODEL</td><td>deepseek-flash<sup>(1)</sup></td><td>deepseek-v4-pro<sup>(2)</sup></td></tr>
|
||||
<tr><td colspan="3">BASE URL (OpenAI Format)</td><td colspan="2">https://api.deepseek.com</td></tr>
|
||||
<tr><td rowspan="6">PRICING<sup>(3)</sup></td><td rowspan="2">1M INPUT TOKENS<br>(CACHE HIT)</td><td>OFF-PEAK</td><td>$0.003</td><td>$0.022</td></tr>
|
||||
<tr><td>PEAK</td><td>$0.006</td><td>$0.044</td></tr>
|
||||
<tr><td rowspan="2">1M INPUT TOKENS<br>(CACHE MISS)</td><td>OFF-PEAK</td><td>$0.15</td><td>$0.66</td></tr>
|
||||
<tr><td>PEAK</td><td>$0.3</td><td>$1.32</td></tr>
|
||||
<tr><td rowspan="2">1M OUTPUT TOKENS</td><td>OFF-PEAK</td><td>$0.6</td><td>$1.98</td></tr>
|
||||
<tr><td>PEAK</td><td>$1.2</td><td>$3.96</td></tr>
|
||||
</table>
|
||||
<p>(1) Use <code>deepseek-flash</code> as the model name. The legacy names <code>deepseek-v4-flash</code> and <code>deepseek-v4-flash-vision-exp</code> are still accepted, but the corresponding models have been retired, their requests are served by the DeepSeek-V4.1-Flash model and billed at the Flash price.</p>
|
||||
<p>(3) Off-peak rates are half of the peak rates. Peak hours are 01:00 - 04:00 and 06:00 - 10:00 UTC, Monday through Friday (all other hours are off-peak).</p>
|
||||
</body></html>`;
|
||||
|
||||
function flashRate(partial?: Partial<ResolvedRate>): ResolvedRate {
|
||||
return {
|
||||
model: "deepseek-flash",
|
||||
aliases: ["deepseek-v4-flash"],
|
||||
currency: "USD",
|
||||
unitTokens: 1_000_000,
|
||||
inputCacheHit: { offPeak: 0.003, peak: 0.006 },
|
||||
inputCacheMiss: { offPeak: 0.15, peak: 0.3 },
|
||||
output: { offPeak: 0.6, peak: 1.2 },
|
||||
peakWindows: [
|
||||
{ startMinutes: 60, endMinutes: 240 },
|
||||
{ startMinutes: 360, endMinutes: 600 },
|
||||
],
|
||||
peakWeekdays: [1, 2, 3, 4, 5],
|
||||
sourceId: "src",
|
||||
sourceName: "DeepSeek",
|
||||
sourceUrl: "https://api-docs.deepseek.com/quick_start/pricing",
|
||||
...partial,
|
||||
};
|
||||
}
|
||||
|
||||
describe("parsePricingDocument DeepSeek HTML", () => {
|
||||
it("reads flash/pro peak and off-peak from the official table", () => {
|
||||
const parsed = parsePricingDocument(DEEPSEEK_HTML, "text/html");
|
||||
expect(parsed.rates.map((r) => r.model)).toEqual([
|
||||
"deepseek-flash",
|
||||
"deepseek-v4-pro",
|
||||
]);
|
||||
const flash = parsed.rates[0];
|
||||
expect(flash.inputCacheHit).toEqual({ offPeak: 0.003, peak: 0.006 });
|
||||
expect(flash.inputCacheMiss).toEqual({ offPeak: 0.15, peak: 0.3 });
|
||||
expect(flash.output).toEqual({ offPeak: 0.6, peak: 1.2 });
|
||||
expect(flash.aliases).toEqual(
|
||||
expect.arrayContaining(["deepseek-v4-flash", "deepseek-v4-flash-vision-exp"]),
|
||||
);
|
||||
expect(flash.peakWeekdays).toEqual([1, 2, 3, 4, 5]);
|
||||
expect(flash.peakWindows).toEqual([
|
||||
{ startMinutes: 60, endMinutes: 240 },
|
||||
{ startMinutes: 360, endMinutes: 600 },
|
||||
]);
|
||||
const pro = parsed.rates[1];
|
||||
expect(pro.inputCacheMiss.offPeak).toBe(0.66);
|
||||
expect(pro.output.peak).toBe(3.96);
|
||||
});
|
||||
});
|
||||
|
||||
describe("parsePricingDocument generic markdown", () => {
|
||||
it("reads model / input / cached / output columns", () => {
|
||||
const md = `
|
||||
| Model | Input | Cached | Output |
|
||||
| --- | --- | --- | --- |
|
||||
| gpt-4o | $2.50 | $1.25 | $10.00 |
|
||||
`;
|
||||
const parsed = parsePricingDocument(md, "text/markdown");
|
||||
expect(parsed.rates).toHaveLength(1);
|
||||
expect(parsed.rates[0]).toMatchObject({
|
||||
model: "gpt-4o",
|
||||
inputCacheMiss: { offPeak: 2.5 },
|
||||
inputCacheHit: { offPeak: 1.25 },
|
||||
output: { offPeak: 10 },
|
||||
});
|
||||
});
|
||||
});
|
||||
|
||||
describe("parsePricingDocument GLM official tables", () => {
|
||||
it("reads Z.AI USD table including Free", () => {
|
||||
const md = `
|
||||
Prices per 1M tokens.
|
||||
|
||||
| Model | Input | Cached Input | Cached Input Storage | Output |
|
||||
| --- | --- | --- | --- | --- |
|
||||
| GLM-5.3-Flash | $0.15 | $0.03 | Limited-time Free | $0.50 |
|
||||
| GLM-5.3 | $1.4 | $0.26 | Limited-time Free | $4.4 |
|
||||
| GLM-4.7-Flash | Free | Free | Free | Free |
|
||||
`;
|
||||
const parsed = parsePricingDocument(md, "text/markdown");
|
||||
const flash = parsed.rates.find((r) => r.model === "GLM-5.3-Flash");
|
||||
expect(flash).toMatchObject({
|
||||
currency: "USD",
|
||||
inputCacheMiss: { offPeak: 0.15 },
|
||||
inputCacheHit: { offPeak: 0.03 },
|
||||
output: { offPeak: 0.5 },
|
||||
});
|
||||
const free = parsed.rates.find((r) => r.model === "GLM-4.7-Flash");
|
||||
expect(free?.inputCacheMiss.offPeak).toBe(0);
|
||||
expect(free?.output.offPeak).toBe(0);
|
||||
});
|
||||
|
||||
it("reads 智谱 CNY 元/百万 Tokens table", () => {
|
||||
const md = `
|
||||
除特别说明外，模型价格统一按照“元/百万 Tokens”展示。
|
||||
|
||||
| 模型名称 | 上下文 | 输入单价（元/百万 Tokens） | 输出单价（元/百万 Tokens） | 缓存存储（元/百万 Tokens/小时） | 缓存命中（元/百万 Tokens） | 输入模态 |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| GLM-5.3 | 1M | 8 | 28 | 限时免费 | 2 | 文本 |
|
||||
| GLM-5.3-Flash | 1M | 0.8 | 2.8 | 限时免费 | 0.23 | 图片、视频、文件、文本 |
|
||||
`;
|
||||
const parsed = parsePricingDocument(md, "text/markdown");
|
||||
const glm = parsed.rates.find((r) => r.model === "GLM-5.3");
|
||||
expect(glm).toMatchObject({
|
||||
currency: "CNY",
|
||||
inputCacheMiss: { offPeak: 8 },
|
||||
inputCacheHit: { offPeak: 2 },
|
||||
output: { offPeak: 28 },
|
||||
});
|
||||
const flash = parsed.rates.find((r) => r.model === "GLM-5.3-Flash");
|
||||
expect(flash?.inputCacheMiss.offPeak).toBe(0.8);
|
||||
expect(flash?.inputCacheHit?.offPeak).toBe(0.23);
|
||||
});
|
||||
|
||||
it("keeps 智谱 CNY even if the HTML also contains $", () => {
|
||||
const html = `
|
||||
<script>var x = "$";</script>
|
||||
<p>模型价格统一按照“元/百万 Tokens”展示。</p>
|
||||
<table>
|
||||
<tr><th>模型名称</th><th>输入单价（元/百万 Tokens）</th><th>输出单价（元/百万 Tokens）</th><th>缓存命中（元/百万 Tokens）</th></tr>
|
||||
<tr><td>GLM-5.3</td><td>8</td><td>28</td><td>2</td></tr>
|
||||
</table>`;
|
||||
const parsed = parsePricingDocument(html, "text/html");
|
||||
expect(parsed.rates[0]).toMatchObject({
|
||||
model: "GLM-5.3",
|
||||
currency: "CNY",
|
||||
inputCacheMiss: { offPeak: 8 },
|
||||
output: { offPeak: 28 },
|
||||
});
|
||||
});
|
||||
});
|
||||
|
||||
describe("parsePricingDocument JSON", () => {
|
||||
it("converts OpenRouter per-token strings to per-million", () => {
|
||||
const parsed = parsePricingDocument(
|
||||
JSON.stringify({
|
||||
data: [
|
||||
{
|
||||
id: "deepseek/deepseek-chat",
|
||||
pricing: { prompt: "0.00000014", completion: "0.00000028" },
|
||||
},
|
||||
],
|
||||
}),
|
||||
"application/json",
|
||||
);
|
||||
expect(parsed.rates[0].model).toBe("deepseek-chat");
|
||||
expect(parsed.rates[0].inputCacheMiss.offPeak).toBeCloseTo(0.14);
|
||||
expect(parsed.rates[0].output.offPeak).toBeCloseTo(0.28);
|
||||
});
|
||||
});
|
||||
|
||||
describe("peak schedule", () => {
|
||||
it("parses DeepSeek UTC weekday windows", () => {
|
||||
const peak = parsePeakSchedule(
|
||||
"Peak hours are 01:00 - 04:00 and 06:00 - 10:00 UTC, Monday through Friday",
|
||||
);
|
||||
expect(peak?.weekdays).toEqual([1, 2, 3, 4, 5]);
|
||||
expect(isPeakAt(new Date("2026-09-14T02:00:00Z"), peak?.windows, peak?.weekdays)).toBe(
|
||||
true,
|
||||
);
|
||||
expect(isPeakAt(new Date("2026-09-13T02:00:00Z"), peak?.windows, peak?.weekdays)).toBe(
|
||||
false,
|
||||
);
|
||||
expect(isPeakAt(new Date("2026-09-14T05:00:00Z"), peak?.windows, peak?.weekdays)).toBe(
|
||||
false,
|
||||
);
|
||||
});
|
||||
|
||||
it("maps legacy flash names", () => {
|
||||
const map = parseLegacyAliases(DEEPSEEK_HTML, ["deepseek-flash"]);
|
||||
expect(map.get("deepseek-flash")).toEqual(
|
||||
expect.arrayContaining(["deepseek-v4-flash"]),
|
||||
);
|
||||
});
|
||||
});
|
||||
|
||||
describe("costForRecord", () => {
|
||||
const rate = flashRate();
|
||||
|
||||
it("uses cache hit/miss and off-peak on weekend", () => {
|
||||
const record: TokenUsageRecord = {
|
||||
id: "1",
|
||||
at: "2026-09-13T02:00:00.000Z",
|
||||
caller: "dictate_agent",
|
||||
model: "deepseek-flash",
|
||||
promptTokens: 1000,
|
||||
completionTokens: 100,
|
||||
totalTokens: 1100,
|
||||
cachedTokens: 800,
|
||||
};
|
||||
expect(splitPromptTokens(record)).toEqual({ cached: 800, miss: 200, output: 100 });
|
||||
const cost = costForRecord(record, rate);
|
||||
expect(cost.currency).toBe("USD");
|
||||
expect(cost.amount).toBeCloseTo((800 * 0.003 + 200 * 0.15 + 100 * 0.6) / 1e6);
|
||||
});
|
||||
|
||||
it("uses peak weekday window", () => {
|
||||
const record: TokenUsageRecord = {
|
||||
id: "2",
|
||||
at: "2026-09-14T02:30:00.000Z",
|
||||
caller: "dictate_agent",
|
||||
model: "deepseek-flash",
|
||||
promptTokens: 1000,
|
||||
completionTokens: 100,
|
||||
totalTokens: 1100,
|
||||
cachedTokens: 800,
|
||||
cacheMissTokens: 200,
|
||||
};
|
||||
const cost = costForRecord(record, rate);
|
||||
expect(cost.amount).toBeCloseTo((800 * 0.006 + 200 * 0.3 + 100 * 1.2) / 1e6);
|
||||
});
|
||||
});
|
||||
|
||||
describe("normalizePricingUrl", () => {
|
||||
it("rejects non-http URLs", async () => {
|
||||
const { normalizePricingUrl, markdownDocFallbacks } = await import("../src/stats/pricing.js");
|
||||
expect(() => normalizePricingUrl("file:///tmp/x")).toThrow(/http/);
|
||||
expect(markdownDocFallbacks("https://docs.bigmodel.cn/cn/guide/start/pricing")).toEqual([
|
||||
"https://docs.bigmodel.cn/cn/guide/start/pricing.md",
|
||||
]);
|
||||
});
|
||||
});
|
||||
|
||||
describe("buildCostReport", () => {
|
||||
beforeAll(() => {
|
||||
process.env.WRITING_AGENT_DATA_DIR = mkdtempSync(
|
||||
path.join(tmpdir(), "wa-pricing-"),
|
||||
);
|
||||
});
|
||||
|
||||
it("groups by model and marks unmatched", () => {
|
||||
const rate = flashRate();
|
||||
const catalog = {
|
||||
byKey: new Map([
|
||||
["deepseek-flash", rate],
|
||||
["deepseek-v4-flash", rate],
|
||||
]),
|
||||
rates: [rate],
|
||||
};
|
||||
const report = buildCostReport(
|
||||
[
|
||||
{
|
||||
id: "a",
|
||||
at: "2026-09-13T08:00:00.000Z",
|
||||
caller: "x",
|
||||
model: "deepseek-v4-flash",
|
||||
promptTokens: 100,
|
||||
completionTokens: 10,
|
||||
totalTokens: 110,
|
||||
cachedTokens: 50,
|
||||
},
|
||||
{
|
||||
id: "b",
|
||||
at: "2026-09-13T08:00:00.000Z",
|
||||
caller: "x",
|
||||
model: "mystery-model",
|
||||
promptTokens: 10,
|
||||
completionTokens: 1,
|
||||
totalTokens: 11,
|
||||
},
|
||||
],
|
||||
catalog,
|
||||
);
|
||||
expect(report.unpricedCalls).toBe(1);
|
||||
expect(report.pricedCalls).toBe(1);
|
||||
expect(report.byModel[0].model).toBe("deepseek-v4-flash");
|
||||
expect(report.byModel[0].cost?.amount).toBeGreaterThan(0);
|
||||
expect(report.byModel.find((r) => r.model === "mystery-model")?.unmatched).toBe(
|
||||
true,
|
||||
);
|
||||
});
|
||||
});
|
||||
Reference in New Issue
Block a user