The best way to control Claude Code, Codex and more from any computer or phone
| 1 | import { describe, expect, it } from "@effect/vitest"; |
| 2 | |
| 3 | import { cursorRateModel } from "@t3tools/provider-cursor/server/accountUsage"; |
| 4 | import type { UsageSpeed } from "@t3tools/provider-core/server/usage"; |
| 5 | import { |
| 6 | cacheSavingsUsd, |
| 7 | createOverrideRateTable, |
| 8 | lookupRate, |
| 9 | parseRateTable, |
| 10 | priceUsage, |
| 11 | } from "./usagePricing.ts"; |
| 12 | |
| 13 | const rate = (input: number, cacheRead?: number) => ({ |
| 14 | input_cost_per_token: input, |
| 15 | output_cost_per_token: input * 5, |
| 16 | ...(cacheRead === undefined ? {} : { cache_read_input_token_cost: cacheRead }), |
| 17 | }); |
| 18 | |
| 19 | describe("usage pricing", () => { |
| 20 | const totals = { |
| 21 | uncachedInputTokens: 1_000_000, |
| 22 | cachedInputTokens: 1_000_000, |
| 23 | cacheCreationTokens: 1_000_000, |
| 24 | outputTokens: 1_000_000, |
| 25 | reasoningTokens: 500_000, |
| 26 | }; |
| 27 | const record = ( |
| 28 | model: string, |
| 29 | reportedCostUsd: number | null = null, |
| 30 | speed: UsageSpeed = "standard", |
| 31 | ) => ({ |
| 32 | model, |
| 33 | totals, |
| 34 | reportedCostUsd, |
| 35 | speed, |
| 36 | }); |
| 37 | |
| 38 | it("uses custom token rates ahead of public and provider-reported costs", () => { |
| 39 | const table = parseRateTable({ "example-model": rate(1) }); |
| 40 | const overrides = createOverrideRateTable({ |
| 41 | "example-model": { |
| 42 | inputCostPerMillionTokens: 2, |
| 43 | outputCostPerMillionTokens: 8, |
| 44 | cacheReadCostPerMillionTokens: 0.5, |
| 45 | cacheWriteCostPerMillionTokens: 3, |
| 46 | }, |
| 47 | }); |
| 48 | |
| 49 | for (const reportedCostUsd of [null, 99]) { |
| 50 | expect(priceUsage(table, record("example-model", reportedCostUsd), overrides)).toMatchObject({ |
| 51 | costUsd: 13.5, |
| 52 | costSource: "modelPriced", |
| 53 | }); |
| 54 | } |
| 55 | expect(cacheSavingsUsd(table, record("example-model"), overrides)).toBe(1.5); |
| 56 | }); |
| 57 | |
| 58 | it("prices Cursor cache savings at the base model rate", () => { |
| 59 | const table = parseRateTable({ |
| 60 | "claude-fable-5-1": rate(10e-6, 1e-6), |
| 61 | "xai/grok-4.7": rate(2e-6, 0.5e-6), |
| 62 | "openrouter/x-ai/grok-4.7": rate(3e-6, 0.5e-6), |
| 63 | }); |
| 64 | const cursorRecord = (model: string) => ({ |
| 65 | ...record(model, 0.25), |
| 66 | rateModel: cursorRateModel(model), |
| 67 | }); |
| 68 | |
| 69 | expect(cacheSavingsUsd(table, cursorRecord("claude-fable-5-1-thinking-high"))).toBeCloseTo(9); |
| 70 | expect(cacheSavingsUsd(table, cursorRecord("cursor-grok-4.7-high-fast"))).toBeCloseTo(1.5); |
| 71 | expect(cacheSavingsUsd(table, cursorRecord("default"))).toBe(0); |
| 72 | expect(priceUsage(table, cursorRecord("grok-4.7-xhigh-fast"))).toMatchObject({ |
| 73 | costUsd: 0.25, |
| 74 | costSource: "providerReported", |
| 75 | }); |
| 76 | }); |
| 77 | |
| 78 | it("falls back to the display model when the rate model is unlisted", () => { |
| 79 | const table = parseRateTable({ "example-model": rate(1e-6, 0.1e-6) }); |
| 80 | const unlisted = { ...record("example-model"), rateModel: "reseller/example-model" }; |
| 81 | |
| 82 | expect(priceUsage(table, unlisted)).toEqual(priceUsage(table, record("example-model"))); |
| 83 | expect(cacheSavingsUsd(table, unlisted)).toBeCloseTo(0.9); |
| 84 | }); |
| 85 | |
| 86 | it("prices unknown models offline and uses input prices for omitted cache rates", () => { |
| 87 | const table = parseRateTable({}); |
| 88 | const overrides = createOverrideRateTable({ |
| 89 | "example-model": { inputCostPerMillionTokens: 2, outputCostPerMillionTokens: 8 }, |
| 90 | }); |
| 91 | |
| 92 | expect(priceUsage(table, record("example-model"), overrides)).toMatchObject({ |
| 93 | costUsd: 14, |
| 94 | costSource: "modelPriced", |
| 95 | }); |
| 96 | expect(cacheSavingsUsd(table, record("example-model"), overrides)).toBe(0); |
| 97 | }); |
| 98 | |
| 99 | it("preserves explicit zero rates and matches only the exact trimmed model ID", () => { |
| 100 | const table = parseRateTable({}); |
| 101 | const overrides = createOverrideRateTable({ |
| 102 | " vendor/example-model[1m] ": { |
| 103 | inputCostPerMillionTokens: 0, |
| 104 | outputCostPerMillionTokens: 0, |
| 105 | }, |
| 106 | }); |
| 107 | expect(priceUsage(table, record(" vendor/example-model[1m] ", 99), overrides)).toMatchObject({ |
| 108 | costUsd: 0, |
| 109 | costSource: "modelPriced", |
| 110 | }); |
| 111 | for (const model of [ |
| 112 | "example-model[1m]", |
| 113 | "vendor/example-model", |
| 114 | "vendor/Example-model[1m]", |
| 115 | "other/example-model[1m]", |
| 116 | ]) { |
| 117 | expect(priceUsage(table, record(model), overrides).costSource).toBe("unpriced"); |
| 118 | expect(priceUsage(table, record(model, 99), overrides)).toEqual({ |
| 119 | costUsd: 99, |
| 120 | costSource: "providerReported", |
| 121 | categoryCostUsd: null, |
| 122 | speedPremiumUsd: 0, |
| 123 | }); |
| 124 | } |
| 125 | }); |
| 126 | |
| 127 | it("prices fast-mode requests at the model's published fast multiple", () => { |
| 128 | const table = parseRateTable({ |
| 129 | "claude-opus-5-5": { ...rate(4e-6, 2e-7), provider_specific_entry: { fast: 2, us: 1.1 } }, |
| 130 | "claude-fable-5-1": { ...rate(1e-5, 2.5e-7), provider_specific_entry: { us: 1.1 } }, |
| 131 | }); |
| 132 | const overrides = createOverrideRateTable({ |
| 133 | "claude-opus-5-5": { inputCostPerMillionTokens: 4, outputCostPerMillionTokens: 20 }, |
| 134 | }); |
| 135 | const cost = (model: string, speed: UsageSpeed, custom?: typeof overrides) => |
| 136 | priceUsage(table, record(model, null, speed), custom).costUsd; |
| 137 | |
| 138 | expect(cost("claude-opus-5-5", "fast")).toBeCloseTo(2 * cost("claude-opus-5-5", "standard")); |
| 139 | expect(cacheSavingsUsd(table, record("claude-opus-5-5", null, "fast"))).toBeCloseTo( |
| 140 | 2 * cacheSavingsUsd(table, record("claude-opus-5-5")), |
| 141 | ); |
| 142 | expect(cost("claude-fable-5-1", "fast")).toBe(cost("claude-fable-5-1", "standard")); |
| 143 | expect(cost("claude-opus-5-5", "fast", overrides)).toBe( |
| 144 | cost("claude-opus-5-5", "standard", overrides), |
| 145 | ); |
| 146 | }); |
| 147 | |
| 148 | it("splits cost by category and prices the speed premium", () => { |
| 149 | const table = parseRateTable({ |
| 150 | "claude-opus-5-5": { |
| 151 | ...rate(4e-6, 4e-7), |
| 152 | cache_creation_input_token_cost: 5e-6, |
| 153 | provider_specific_entry: { fast: 2 }, |
| 154 | }, |
| 155 | }); |
| 156 | const split = (input: number, cacheRead: number, cacheWrite: number, output: number) => ({ |
| 157 | input: expect.closeTo(input), |
| 158 | cacheRead: expect.closeTo(cacheRead), |
| 159 | cacheWrite: expect.closeTo(cacheWrite), |
| 160 | output: expect.closeTo(output), |
| 161 | }); |
| 162 | |
| 163 | expect(priceUsage(table, record("claude-opus-5-5", null, "fast"))).toEqual({ |
| 164 | costUsd: expect.closeTo(58.8), |
| 165 | costSource: "modelPriced", |
| 166 | categoryCostUsd: split(8, 0.8, 10, 40), |
| 167 | speedPremiumUsd: expect.closeTo(29.4), |
| 168 | }); |
| 169 | expect(priceUsage(table, record("claude-opus-5-5", 29.4, "fast"))).toEqual({ |
| 170 | costUsd: 29.4, |
| 171 | costSource: "providerReported", |
| 172 | categoryCostUsd: split(4, 0.4, 5, 20), |
| 173 | speedPremiumUsd: expect.closeTo(14.7), |
| 174 | }); |
| 175 | expect(priceUsage(table, record("unknown-model", 29.4, "fast"))).toMatchObject({ |
| 176 | costUsd: 29.4, |
| 177 | categoryCostUsd: null, |
| 178 | speedPremiumUsd: 0, |
| 179 | }); |
| 180 | }); |
| 181 | |
| 182 | it("prices Codex priority and ultrafast requests at their published tier rates", () => { |
| 183 | const table = parseRateTable({ |
| 184 | "gpt-6-astra": { |
| 185 | ...rate(1e-5, 1e-6), |
| 186 | input_cost_per_token_priority: 2e-5, |
| 187 | output_cost_per_token_priority: 1e-4, |
| 188 | cache_read_input_token_cost_priority: 2e-6, |
| 189 | input_cost_per_token_ultrafast: 6e-5, |
| 190 | output_cost_per_token_ultrafast: 3e-4, |
| 191 | }, |
| 192 | "gpt-6-sol": rate(2e-6, 2e-7), |
| 193 | }); |
| 194 | const cost = (model: string, speed: UsageSpeed) => |
| 195 | priceUsage(table, record(model, null, speed)).costUsd; |
| 196 | const standard = cost("gpt-6-astra", "standard"); |
| 197 | |
| 198 | expect(cost("gpt-6-astra", "fast")).toBeCloseTo(2 * standard); |
| 199 | expect(cost("gpt-6-astra", "ultrafast")).toBeCloseTo(6 * standard); |
| 200 | expect(cacheSavingsUsd(table, record("gpt-6-astra", null, "ultrafast"))).toBeCloseTo( |
| 201 | 6 * cacheSavingsUsd(table, record("gpt-6-astra")), |
| 202 | ); |
| 203 | expect(cost("gpt-6-sol", "ultrafast")).toBe(cost("gpt-6-sol", "standard")); |
| 204 | }); |
| 205 | |
| 206 | it("prices a request above the model's long-context threshold at its above-tier rates", () => { |
| 207 | const table = parseRateTable({ |
| 208 | "claude-haiku-5-5": { |
| 209 | input_cost_per_token: 1e-6, |
| 210 | output_cost_per_token: 5e-6, |
| 211 | cache_read_input_token_cost: 1e-7, |
| 212 | cache_creation_input_token_cost: 1.25e-6, |
| 213 | input_cost_per_token_above_100k_tokens: 2e-6, |
| 214 | output_cost_per_token_above_100k_tokens: 7.5e-6, |
| 215 | cache_read_input_token_cost_above_100k_tokens: 2e-7, |
| 216 | cache_creation_input_token_cost_above_100k_tokens: 2.5e-6, |
| 217 | }, |
| 218 | "gpt-6-sol": { |
| 219 | ...rate(2e-6, 2e-7), |
| 220 | input_cost_per_token_priority: 4e-6, |
| 221 | output_cost_per_token_priority: 2e-5, |
| 222 | cache_read_input_token_cost_priority: 4e-7, |
| 223 | input_cost_per_token_above_272k_tokens_priority: 8e-6, |
| 224 | output_cost_per_token_above_272k_tokens_priority: 3e-5, |
| 225 | cache_read_input_token_cost_above_272k_tokens_priority: 8e-7, |
| 226 | }, |
| 227 | }); |
| 228 | const longRequest = ( |
| 229 | model: string, |
| 230 | speed: UsageSpeed, |
| 231 | totals: { uncached: number; cacheRead: number; cacheWrite: number; output: number }, |
| 232 | ) => ({ |
| 233 | model, |
| 234 | speed, |
| 235 | reportedCostUsd: null, |
| 236 | totals: { |
| 237 | uncachedInputTokens: totals.uncached, |
| 238 | cachedInputTokens: totals.cacheRead, |
| 239 | cacheCreationTokens: totals.cacheWrite, |
| 240 | outputTokens: totals.output, |
| 241 | reasoningTokens: 0, |
| 242 | }, |
| 243 | }); |
| 244 | |
| 245 | // The Claude prompt is input + cache read + cache write: 120K, above the 100K tier. |
| 246 | expect |
| 247 | .soft( |
| 248 | priceUsage( |
| 249 | table, |
| 250 | longRequest("claude-haiku-5-5", "standard", { |
| 251 | uncached: 10_000, |
| 252 | cacheRead: 100_000, |
| 253 | cacheWrite: 10_000, |
| 254 | output: 1_000, |
| 255 | }), |
| 256 | ).costUsd, |
| 257 | ) |
| 258 | .toBeCloseTo(10_000 * 2e-6 + 100_000 * 2e-7 + 10_000 * 2.5e-6 + 1_000 * 7.5e-6, 12); |
| 259 | expect( |
| 260 | priceUsage( |
| 261 | table, |
| 262 | longRequest("gpt-6-sol", "fast", { |
| 263 | uncached: 20_000, |
| 264 | cacheRead: 280_000, |
| 265 | cacheWrite: 0, |
| 266 | output: 1_000, |
| 267 | }), |
| 268 | ).costUsd, |
| 269 | ).toBeCloseTo(20_000 * 8e-6 + 280_000 * 8e-7 + 1_000 * 3e-5, 12); |
| 270 | }); |
| 271 | |
| 272 | it("bills the long-context tier only above its threshold, filling missing tier fields from base", () => { |
| 273 | const table = parseRateTable({ |
| 274 | "gpt-6-sol": { |
| 275 | input_cost_per_token: 2e-6, |
| 276 | output_cost_per_token: 1e-5, |
| 277 | cache_read_input_token_cost: 2e-7, |
| 278 | input_cost_per_token_priority: 4e-6, |
| 279 | output_cost_per_token_priority: 2e-5, |
| 280 | input_cost_per_token_above_272k_tokens: 4e-6, |
| 281 | }, |
| 282 | }); |
| 283 | const cost = (prompt: number, speed: UsageSpeed = "standard") => |
| 284 | priceUsage(table, { |
| 285 | model: "gpt-6-sol", |
| 286 | speed, |
| 287 | reportedCostUsd: null, |
| 288 | totals: { |
| 289 | uncachedInputTokens: prompt - 200_000, |
| 290 | cachedInputTokens: 200_000, |
| 291 | cacheCreationTokens: 0, |
| 292 | outputTokens: 1_000, |
| 293 | reasoningTokens: 0, |
| 294 | }, |
| 295 | }).costUsd; |
| 296 | |
| 297 | expect(cost(272_000)).toBeCloseTo(72_000 * 2e-6 + 200_000 * 2e-7 + 1_000 * 1e-5, 12); |
| 298 | // Output keeps the base rate; cache reads keep the base discount relative to input. |
| 299 | expect(cost(272_001)).toBeCloseTo(72_001 * 4e-6 + 200_000 * 4e-7 + 1_000 * 1e-5, 12); |
| 300 | // Without a priority long-context rate, priority requests keep the priority rate. |
| 301 | expect(cost(272_001, "fast")).toBeCloseTo(72_001 * 4e-6 + 200_000 * 4e-7 + 1_000 * 2e-5, 12); |
| 302 | }); |
| 303 | |
| 304 | it("keeps the canonical Fable rate separate from DeepInfra in either order", () => { |
| 305 | const canonical = ["claude-fable-5", rate(1e-5, 1e-6)] as const; |
| 306 | const deepInfra = ["deepinfra/anthropic/claude-fable-5", rate(1e-5)] as const; |
| 307 | |
| 308 | for (const entries of [ |
| 309 | [canonical, deepInfra], |
| 310 | [deepInfra, canonical], |
| 311 | ]) { |
| 312 | const table = parseRateTable(Object.fromEntries(entries)); |
| 313 | |
| 314 | expect(lookupRate(table, "claude-fable-5")?.cacheReadCostPerToken).toBe(1e-6); |
| 315 | expect(lookupRate(table, "deepinfra/anthropic/claude-fable-5")?.cacheReadCostPerToken).toBe( |
| 316 | 1e-5, |
| 317 | ); |
| 318 | expect(lookupRate(table, "other/claude-fable-5")).toBeNull(); |
| 319 | } |
| 320 | }); |
| 321 | |
| 322 | it("prices a bracketed context-tier variant at the base model's rate", () => { |
| 323 | const table = parseRateTable({ "claude-fable-5-1": rate(1e-5, 2.5e-7) }); |
| 324 | |
| 325 | expect(lookupRate(table, "claude-fable-5-1[1m]")).toEqual( |
| 326 | lookupRate(table, "claude-fable-5-1"), |
| 327 | ); |
| 328 | expect(lookupRate(table, "anthropic/Claude-Fable-5-1[1m]")).toBeNull(); |
| 329 | }); |
| 330 | |
| 331 | it("adds a bare alias when every qualified entry has the same rate", () => { |
| 332 | const table = parseRateTable({ |
| 333 | "provider-a/example-model": rate(1), |
| 334 | "provider-b/example-model": rate(1), |
| 335 | }); |
| 336 | |
| 337 | expect(lookupRate(table, "example-model")).toEqual( |
| 338 | lookupRate(table, "provider-a/example-model"), |
| 339 | ); |
| 340 | }); |
| 341 | |
| 342 | it("keeps a bare alias when a reseller entry omits the fast tier", () => { |
| 343 | const table = parseRateTable({ |
| 344 | "provider-a/example-model": { |
| 345 | ...rate(1), |
| 346 | input_cost_per_token_priority: 2, |
| 347 | output_cost_per_token_priority: 10, |
| 348 | input_cost_per_token_above_272k_tokens: 2, |
| 349 | input_cost_per_token_above_272k_tokens_priority: 4, |
| 350 | output_cost_per_token_above_272k_tokens_priority: 20, |
| 351 | }, |
| 352 | "provider-b/example-model": { ...rate(1), input_cost_per_token_above_272k_tokens: 2 }, |
| 353 | "provider-c/example-model": rate(1), |
| 354 | }); |
| 355 | |
| 356 | expect(lookupRate(table, "example-model")).toEqual( |
| 357 | lookupRate(table, "provider-a/example-model"), |
| 358 | ); |
| 359 | }); |
| 360 | |
| 361 | it("leaves an ambiguous bare name unpriced", () => { |
| 362 | const table = parseRateTable({ |
| 363 | "provider-a/example-model": rate(1), |
| 364 | "provider-b/example-model": rate(3), |
| 365 | }); |
| 366 | |
| 367 | expect(lookupRate(table, "provider-a/example-model")?.inputCostPerToken).toBe(1); |
| 368 | expect(lookupRate(table, "provider-b/example-model")?.inputCostPerToken).toBe(3); |
| 369 | expect(lookupRate(table, "example-model")).toBeNull(); |
| 370 | }); |
| 371 | |
| 372 | it("prices a bare name at the model maker's rate when a reseller marks it up", () => { |
| 373 | const reseller = { "kyma/gemini-4-argon": rate(5.4e-6) }; |
| 374 | const makers = { |
| 375 | "gemini/gemini-4-argon": rate(2e-6, 1e-7), |
| 376 | "vertex_ai/gemini-4-argon": rate(2e-6, 1e-7), |
| 377 | }; |
| 378 | |
| 379 | for (const document of [ |
| 380 | { ...reseller, ...makers }, |
| 381 | { ...makers, ...reseller }, |
| 382 | ]) { |
| 383 | const table = parseRateTable(document); |
| 384 | expect(lookupRate(table, "gemini-4-argon")).toEqual( |
| 385 | lookupRate(table, "gemini/gemini-4-argon"), |
| 386 | ); |
| 387 | } |
| 388 | }); |
| 389 | |
| 390 | it("prices a bare name listed only under Bedrock IDs at the base anthropic rate", () => { |
| 391 | const bedrock = (input: number, output: number, cacheRead: number, cacheCreation: number) => ({ |
| 392 | input_cost_per_token: input, |
| 393 | output_cost_per_token: output, |
| 394 | cache_read_input_token_cost: cacheRead, |
| 395 | cache_creation_input_token_cost: cacheCreation, |
| 396 | }); |
| 397 | const base = bedrock(8e-6, 4e-5, 2e-7, 1e-5); |
| 398 | const regional = bedrock(8.8e-6, 4.4e-5, 2.2e-7, 1.1e-5); |
| 399 | const govCloud = bedrock(9.6e-6, 4.8e-5, 2.4e-7, 1.2e-5); |
| 400 | const rows = [ |
| 401 | ["anthropic.claude-mythos-5-2", base], |
| 402 | ["global.anthropic.claude-mythos-5-2", base], |
| 403 | ["us.anthropic.claude-mythos-5-2", regional], |
| 404 | ["eu.anthropic.claude-mythos-5-2", regional], |
| 405 | ["us-gov.anthropic.claude-mythos-5-2", govCloud], |
| 406 | ["bedrock/us-gov-east-1/anthropic.claude-mythos-5-2", govCloud], |
| 407 | ["bedrock/us-gov-west-1/anthropic.claude-mythos-5-2", govCloud], |
| 408 | ] as const; |
| 409 | |
| 410 | for (const entries of [rows, rows.toReversed()]) { |
| 411 | const table = parseRateTable(Object.fromEntries(entries)); |
| 412 | expect(lookupRate(table, "claude-mythos-5-2")).toMatchObject({ |
| 413 | inputCostPerToken: 8e-6, |
| 414 | outputCostPerToken: 4e-5, |
| 415 | cacheReadCostPerToken: 2e-7, |
| 416 | cacheCreationCostPerToken: 1e-5, |
| 417 | }); |
| 418 | } |
| 419 | }); |
| 420 | |
| 421 | it("leaves a bare name unpriced when model makers' own entries disagree", () => { |
| 422 | const table = parseRateTable({ |
| 423 | "openai/example-model": rate(1), |
| 424 | "xai/example-model": rate(3), |
| 425 | "kyma/example-model": rate(1), |
| 426 | }); |
| 427 | |
| 428 | expect(lookupRate(table, "example-model")).toBeNull(); |
| 429 | }); |
| 430 | }); |