DeepSeek V4.1 Flash Provider Benchmark

[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8,"_9":10},"actionData","errors","root",{"_471":472,"_473":474},"routes/_app.x.$handle.$name",{"_111":110,"_38":52,"_119":97,"_33":120,"_13":14,"_121":-5,"_11":122,"_123":124,"_125":126,"_127":128,"_129":130,"_26":-5,"_131":132,"_133":134,"_135":31,"_136":137,"_138":66,"_139":-5,"_140":-5,"_141":142},"routes/_app.x.$handle.$name.branch.$branch.version.$version.diff",{"_11":12,"_13":14,"_15":16,"_17":18,"_19":-5},"project",{"_50":51,"_38":52,"_45":53,"_54":-5,"_55":31,"_56":57,"_58":57,"_59":60,"_61":-5,"_19":62,"_63":64,"_65":66,"_67":-5,"_68":-5,"_69":70,"_71":72,"_73":74,"_75":76,"_77":78,"_79":-5,"_80":81,"_82":81,"_83":84,"_85":84,"_86":84,"_87":81,"_88":81,"_89":90,"_91":92,"_93":94,"_95":31},"version",15,"changes",[25],"aiChatInfo",{"_20":21,"_22":-5,"_23":-5,"_24":-5},"userId","aiChatMessageId","5c56d3ef-9c4e-4adc-846f-493e7137e643","aiChatThreadId","aiChatThreadUserId","aiChatThreadTitle",{"_26":27,"_28":29,"_30":31,"_32":31,"_33":34,"_35":34},"file",{"_36":37,"_38":39,"_40":41,"_42":44,"_45":46,"_47":14,"_48":49,"_30":31},"previous",{"_36":37,"_38":39,"_40":41,"_42":43},"deleted",false,"deletedHere","path","benchmark/execute.ts","latestPath","fileId","01a08da5-899f-73ab-baf1-82bcd375816a","name","execute.ts","fileType","script","content","import { createGateway } from \"npm:@ai-sdk/gateway@3.0.134\";\nimport type { LanguageModel } from \"npm:ai@6.0.168\";\nimport {\n createBenchmarkRun,\n findResumableRun,\n finishBenchmarkRun,\n getBenchmarkRun,\n getRecordedAttempts,\n saveMeasurement,\n} from \"../db/queries.ts\";\nimport type {\n BenchmarkConfig,\n BenchmarkRunDetail,\n GatewayEndpoint,\n} from \"../shared/types.ts\";\nimport { runAttempt } from \"./attempt.ts\";\nimport { fetchGatewayEndpoints } from \"./endpoints.ts\";\nimport {\n BenchmarkLockLostError,\n type EnsureBenchmarkLease,\n type SkippedBenchmarkExecution,\n withBenchmarkExecutionLease,\n} from \"./lease.ts\";\n\ninterface ProviderRunContext {\n runId: string;\n config: BenchmarkConfig;\n model: LanguageModel;\n recorded: Set;\n ensureLease: EnsureBenchmarkLease;\n}\n\nfunction errorMessage(error: unknown): string {\n return error instanceof Error ? error.message : String(error);\n}\n\nasync function runProvider(\n endpoint: GatewayEndpoint,\n context: ProviderRunContext,\n): Promise {\n for (\n let attempt = 1;\n attempt <= context.config.runsPerProvider;\n attempt++\n ) {\n const key = `${endpoint.provider_name}:${attempt}`;\n if (context.recorded.has(key)) {\n console.log(`Skipping saved attempt ${key}`);\n continue;\n }\n\n console.log(\n `Running ${endpoint.provider_name} attempt ${attempt}/${context.config.runsPerProvider}`,\n );\n const measurement = await runAttempt({\n runId: context.runId,\n config: context.config,\n endpoint,\n attempt,\n model: context.model,\n });\n await context.ensureLease();\n await saveMeasurement(measurement);\n context.recorded.add(key);\n console.log(\n `${endpoint.provider_name} attempt ${attempt}: ${measurement.status}`,\n );\n }\n}\n\nasync function runProviders(\n endpoints: GatewayEndpoint[],\n context: ProviderRunContext,\n): Promise {\n let nextProvider = 0;\n\n async function worker(): Promise {\n while (true) {\n const endpoint = endpoints[nextProvider++];\n if (!endpoint) return;\n await runProvider(endpoint, context);\n }\n }\n\n const workerCount = Math.max(\n 1,\n Math.min(context.config.providerConcurrency, endpoints.length),\n );\n await Promise.all(\n Array.from({ length: workerCount }, () => worker()),\n );\n}\n\nasync function executeBenchmarkUnlocked(\n config: BenchmarkConfig,\n ensureLease: EnsureBenchmarkLease,\n): Promise {\n const apiKey = Deno.env.get(\"VERCEL_AI_GATEWAY_API_KEY\");\n if (!apiKey) {\n throw new Error(\"Missing VERCEL_AI_GATEWAY_API_KEY environment variable\");\n }\n\n const configJson = JSON.stringify(config);\n const resumableRunId = await findResumableRun(configJson);\n let endpoints: GatewayEndpoint[];\n\n if (resumableRunId) {\n const existing = await getBenchmarkRun(resumableRunId);\n if (!existing) {\n throw new Error(`Resumable benchmark ${resumableRunId} was not found`);\n }\n endpoints = existing.endpointSnapshot;\n console.log(`Resuming benchmark ${resumableRunId}`);\n } else {\n endpoints = [\n ...await fetchGatewayEndpoints(config.modelId),\n ...GATEWAY_STRATEGIES,\n ];\n }\n\n if (config.providers?.length) {\n const allowed = new Set(config.providers);\n endpoints = endpoints.filter((endpoint) =>\n endpoint.kind === \"strategy\" || allowed.has(endpoint.provider_name)\n );\n }\n if (endpoints.length === 0) {\n throw new Error(\"No matching providers found in the endpoint catalogue\");\n }\n\n await ensureLease();\n\n const runId = resumableRunId ?? crypto.randomUUID();\n if (!resumableRunId) {\n await createBenchmarkRun({ id: runId, config, endpoints });\n console.log(`Created benchmark ${runId}`);\n }\n\n const gateway = createGateway({ apiKey });\n const model = gateway(config.modelId);\n\n try {\n await runProviders(endpoints, {\n runId,\n config,\n model,\n recorded: await getRecordedAttempts(runId),\n ensureLease,\n });\n await ensureLease(true);\n } catch (error) {\n if (error instanceof BenchmarkLockLostError) throw error;\n await finishBenchmarkRun(runId, \"failed\", errorMessage(error));\n throw error;\n }\n\n await finishBenchmarkRun(runId, \"completed\");\n const detail = await getBenchmarkRun(runId);\n if (!detail) {\n throw new Error(`Completed benchmark ${runId} was not found`);\n }\n\n console.log(\n JSON.stringify({\n runId,\n status: detail.status,\n providers: detail.providerCount,\n successfulMeasurements: detail.successfulMeasurements,\n failedMeasurements: detail.failedMeasurements,\n }),\n );\n return detail;\n}\n\nexport function executeBenchmark(\n config: BenchmarkConfig,\n): Promise {\n return withBenchmarkExecutionLease(\n config.modelId,\n (ensureLease) => executeBenchmarkUnlocked(config, ensureLease),\n );\n}\n","import { createGateway } from \"npm:@ai-sdk/gateway@3.0.134\";\nimport type { LanguageModel } from \"npm:ai@6.0.168\";\nimport {\n createBenchmarkRun,\n findResumableRun,\n finishBenchmarkRun,\n getBenchmarkRun,\n getRecordedAttempts,\n saveMeasurement,\n} from \"../db/queries.ts\";\nimport type {\n BenchmarkConfig,\n BenchmarkRunDetail,\n GatewayEndpoint,\n} from \"../shared/types.ts\";\nimport { runAttempt } from \"./attempt.ts\";\nimport { fetchGatewayEndpoints } from \"./endpoints.ts\";\nimport { GATEWAY_STRATEGIES } from \"./strategies.ts\";\nimport {\n BenchmarkLockLostError,\n type EnsureBenchmarkLease,\n type SkippedBenchmarkExecution,\n withBenchmarkExecutionLease,\n} from \"./lease.ts\";\n\ninterface ProviderRunContext {\n runId: string;\n config: BenchmarkConfig;\n model: LanguageModel;\n recorded: Set;\n ensureLease: EnsureBenchmarkLease;\n}\n\nfunction errorMessage(error: unknown): string {\n return error instanceof Error ? error.message : String(error);\n}\n\nasync function runProvider(\n endpoint: GatewayEndpoint,\n context: ProviderRunContext,\n): Promise {\n for (\n let attempt = 1;\n attempt <= context.config.runsPerProvider;\n attempt++\n ) {\n const key = `${endpoint.provider_name}:${attempt}`;\n if (context.recorded.has(key)) {\n console.log(`Skipping saved attempt ${key}`);\n continue;\n }\n\n console.log(\n `Running ${endpoint.provider_name} attempt ${attempt}/${context.config.runsPerProvider}`,\n );\n const measurement = await runAttempt({\n runId: context.runId,\n config: context.config,\n endpoint,\n attempt,\n model: context.model,\n });\n await context.ensureLease();\n await saveMeasurement(measurement);\n context.recorded.add(key);\n console.log(\n `${endpoint.provider_name} attempt ${attempt}: ${measurement.status}`,\n );\n }\n}\n\nasync function runProviders(\n endpoints: GatewayEndpoint[],\n context: ProviderRunContext,\n): Promise {\n let nextProvider = 0;\n\n async function worker(): Promise {\n while (true) {\n const endpoint = endpoints[nextProvider++];\n if (!endpoint) return;\n await runProvider(endpoint, context);\n }\n }\n\n const workerCount = Math.max(\n 1,\n Math.min(context.config.providerConcurrency, endpoints.length),\n );\n await Promise.all(\n Array.from({ length: workerCount }, () => worker()),\n );\n}\n\nasync function executeBenchmarkUnlocked(\n config: BenchmarkConfig,\n ensureLease: EnsureBenchmarkLease,\n): Promise {\n const apiKey = Deno.env.get(\"VERCEL_AI_GATEWAY_API_KEY\");\n if (!apiKey) {\n throw new Error(\"Missing VERCEL_AI_GATEWAY_API_KEY environment variable\");\n }\n\n const configJson = JSON.stringify(config);\n const resumableRunId = await findResumableRun(configJson);\n let endpoints: GatewayEndpoint[];\n\n if (resumableRunId) {\n const existing = await getBenchmarkRun(resumableRunId);\n if (!existing) {\n throw new Error(`Resumable benchmark ${resumableRunId} was not found`);\n }\n endpoints = existing.endpointSnapshot;\n console.log(`Resuming benchmark ${resumableRunId}`);\n } else {\n endpoints = [\n ...await fetchGatewayEndpoints(config.modelId),\n ...GATEWAY_STRATEGIES,\n ];\n }\n\n if (config.providers?.length) {\n const allowed = new Set(config.providers);\n endpoints = endpoints.filter((endpoint) =>\n endpoint.kind === \"strategy\" || allowed.has(endpoint.provider_name)\n );\n }\n if (endpoints.length === 0) {\n throw new Error(\"No matching providers found in the endpoint catalogue\");\n }\n\n await ensureLease();\n\n const runId = resumableRunId ?? crypto.randomUUID();\n if (!resumableRunId) {\n await createBenchmarkRun({ id: runId, config, endpoints });\n console.log(`Created benchmark ${runId}`);\n }\n\n const gateway = createGateway({ apiKey });\n const model = gateway(config.modelId);\n\n try {\n await runProviders(endpoints, {\n runId,\n config,\n model,\n recorded: await getRecordedAttempts(runId),\n ensureLease,\n });\n await ensureLease(true);\n } catch (error) {\n if (error instanceof BenchmarkLockLostError) throw error;\n await finishBenchmarkRun(runId, \"failed\", errorMessage(error));\n throw error;\n }\n\n await finishBenchmarkRun(runId, \"completed\");\n const detail = await getBenchmarkRun(runId);\n if (!detail) {\n throw new Error(`Completed benchmark ${runId} was not found`);\n }\n\n console.log(\n JSON.stringify({\n runId,\n status: detail.status,\n providers: detail.providerCount,\n successfulMeasurements: detail.successfulMeasurements,\n failedMeasurements: detail.failedMeasurements,\n }),\n );\n return detail;\n}\n\nexport function executeBenchmark(\n config: BenchmarkConfig,\n): Promise {\n return withBenchmarkExecutionLease(\n config.modelId,\n (ensureLease) => executeBenchmarkUnlocked(config, ensureLease),\n );\n}\n","createdAt",["D",1789484590148],"branchVersion","branchId","0cb5a767-9630-4119-a265-8c9f5f633c9a","id","1ea4658b-1977-4a2c-8356-9d6ab6609397","deepseek-v41-flash-benchmark",["D",1789082831202],"deletedAt","legacyEnvVars","privacy","public","httpPrivacy","description","Run and compare AI Gateway providers for deepseek-v4.1-flash","imageUrl","a0bf3b31-15a5-4d5c-880e-4b1e22c9bc18","pinnedFileId","01a08da5-899f-73ab-baf1-f0227ea1ff9b","pinPreview",true,"pinMigrated","screenshotVersion","likeCount",0,"orgId","e81e7335-6d4c-4764-a902-d0962bbfa9f7","tursoName","1ea4658b19774a2c83569d6ab6609397","tursoHostname","1ea4658b19774a2c83569d6ab6609397-val-town.aws-us-east-2.turso.io","tursoMeasuredAt",["D",1789518676100],"pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm","user",{"_111":110,"_118":114},"org",{"_111":110,"_112":-5,"_113":114,"_19":62,"_115":116,"_117":31},"branch",{"_13":96,"_38":97,"_50":49,"_98":51,"_99":100,"_101":102,"_103":57,"_104":105,"_106":107,"_108":52,"_109":110},"isOwned",81,"main","projectId","forkedBranchId","24cb14b4-1851-4548-adc6-43a574860a3d","forkedBranchVersion",322,"forkedProjectPrivacy","forkedProjectId","d055324c-28f9-4529-80dd-e9bbfbdbeb65","forkedProjectOrgId","ba7bb5fb-1abb-4ffb-8f3d-7c244782e5f7","projectName","userHandle","stevekrouse","handle","displayName","avatar","https://images.clerk.dev/uploaded/img_2PqHa2Gsy93xQrjh2w78Xu0cChW.jpeg","tier","pro","canRestrictVals","profileImageUrl","branchName","","adoptOrgId",{"_50":51,"_38":52,"_45":466,"_54":-5,"_55":31,"_56":57,"_58":57,"_59":60,"_61":-5,"_19":62,"_63":64,"_65":66,"_67":-5,"_68":-5,"_69":70,"_71":72,"_73":74,"_75":76,"_77":467,"_79":-5,"_80":81,"_82":81,"_83":84,"_85":84,"_86":84,"_87":81,"_88":81,"_89":468,"_91":469,"_93":470,"_95":31},"tree",[217,227,232,246,248,249,250],"files",[217,218,219,220,221,222,223,224,225,226,227,228,229,230,231,232,233,234,235,236,237,238,239,240,241,242,243,244,245,246,247,248,249,250],"counts",{"_207":70,"_208":164,"_209":70,"_210":211,"_212":213},"contents",[154,155,156,157,158,159,160],"upstreamProject",{"_50":105,"_38":150,"_59":151,"_61":-5,"_56":57,"_89":152},"upstreamBranch",{"_50":100,"_45":143,"_144":145,"_38":97,"_98":105,"_13":146,"_99":-5,"_101":-5,"_147":148,"_54":-5,"_95":31,"_56":57,"_109":149,"_71":107,"_108":150},"emptyEnvVars","flags",[],"hasHttpFiles","grants","orgs","customDomains",{},["D",1787848294548],"updatedAt",["D",1789083925414],338,"lockfileKey","1789083928454-esm.town_v_valdottown_glm53-flash-benchmark@338-main","valdottown","glm53-flash-benchmark","Run and compare AI Gateway providers",{"_111":149,"_118":153},"https://img.clerk.com/eyJ0eXBlIjoicHJveHkiLCJzcmMiOiJodHRwczovL2ltYWdlcy5jbGVyay5kZXYvdXBsb2FkZWQvaW1nXzJyWnQ5cFE0c3pleVRCSEVTWmxzaXJsTUlGciJ9",{"_50":202,"_38":203,"_36":204,"_163":-5,"_48":49,"_47":164,"_165":66,"_45":205,"_13":70,"_40":190,"_42":-5,"_30":31,"_169":-5,"_33":206},{"_50":197,"_38":198,"_36":199,"_163":-5,"_48":49,"_47":164,"_165":66,"_45":200,"_13":70,"_40":190,"_42":-5,"_30":31,"_169":-5,"_33":201},{"_50":192,"_38":193,"_36":194,"_163":-5,"_48":49,"_47":164,"_165":66,"_45":195,"_13":70,"_40":190,"_42":-5,"_30":31,"_169":-5,"_33":196},{"_50":186,"_38":187,"_36":188,"_163":-5,"_48":49,"_47":164,"_165":66,"_45":189,"_13":70,"_40":190,"_42":-5,"_30":31,"_169":-5,"_33":191},{"_50":179,"_38":180,"_36":181,"_163":-5,"_48":49,"_47":182,"_165":31,"_45":183,"_13":70,"_40":26,"_42":184,"_30":31,"_169":-5,"_33":185},{"_50":172,"_38":173,"_36":174,"_163":-5,"_48":49,"_47":175,"_165":31,"_45":176,"_13":164,"_40":26,"_42":177,"_30":31,"_169":-5,"_33":178},{"_50":161,"_38":162,"_36":64,"_163":-5,"_48":49,"_47":164,"_165":66,"_45":166,"_13":70,"_40":167,"_42":168,"_30":31,"_169":170,"_33":171},"8ad4c023-63e7-4774-ae26-ef9aa9990c1f","index.ts","parentId",1,"latest",["D",1789082831202],"http","import { parseVal, serveImmutableFile } from \"https://esm.town/v/std/utils/index.ts\";\nimport { Hono } from \"npm:hono@4\";\nimport { FULL_BENCHMARK_CONFIG } from \"./benchmark/config.ts\";\nimport { listMeasurements } from \"./db/measurements.ts\";\nimport { getBenchmarkResults } from \"./db/results.ts\";\nimport { Root } from \"./frontend/root.tsx\";\nimport type {\n MeasurementPublicStatus,\n MeasurementSortDirection,\n MeasurementSortKey,\n} from \"./shared/types.ts\";\n\nconst MEASUREMENT_SORT_KEYS = new Set([\n \"runStartedAt\",\n \"provider\",\n \"attempt\",\n \"status\",\n \"startedAt\",\n \"ttftMs\",\n \"durationMs\",\n \"tokensPerSecond\",\n \"inputTokens\",\n \"outputTokens\",\n \"cacheReadTokens\",\n \"cacheWriteTokens\",\n \"costUsd\",\n \"message\",\n]);\nconst app = new Hono();\n\nfunction measurementSortKey(value: string | undefined): MeasurementSortKey {\n const key = value as MeasurementSortKey | undefined;\n return key && MEASUREMENT_SORT_KEYS.has(key) ? key : \"startedAt\";\n}\n\nfunction measurementStatus(\n value: string | undefined,\n): MeasurementPublicStatus | undefined {\n return value === \"success\" || value === \"error\" ? value : undefined;\n}\n\nfunction positiveInteger(\n value: string | undefined,\n fallback: number,\n maximum: number,\n): number {\n const parsed = Number(value);\n return Number.isInteger(parsed) && parsed > 0\n ? Math.min(parsed, maximum)\n : fallback;\n}\n\nfunction nonNegativeInteger(\n value: string | undefined,\n fallback: number,\n maximum: number,\n): number {\n const parsed = Number(value);\n return Number.isSafeInteger(parsed) && parsed >= 0\n ? Math.min(parsed, maximum)\n : fallback;\n}\n\napp.get(\"/\", (c) => c.html(Root()));\napp.get(\"/measurements\", (c) => c.html(Root()));\n\napp.get(\"/__immutable/*\", (c) => serveImmutableFile(c.req.path));\n\napp.get(\"/api/results\", async (c) => {\n const results = await getBenchmarkResults(FULL_BENCHMARK_CONFIG.modelId);\n c.header(\"Cache-Control\", \"no-store\");\n c.header(\"X-Content-Type-Options\", \"nosniff\");\n return c.json({ results });\n});\n\napp.get(\"/api/measurements\", async (c) => {\n const attemptValue = c.req.query(\"attempt\");\n const attempt = positiveInteger(attemptValue, 0, 100);\n const direction: MeasurementSortDirection =\n c.req.query(\"direction\") === \"asc\" ? \"asc\" : \"desc\";\n\n const measurements = await listMeasurements({\n modelId: FULL_BENCHMARK_CONFIG.modelId,\n provider: c.req.query(\"provider\") || undefined,\n status: measurementStatus(c.req.query(\"status\")),\n attempt: attempt > 0 ? attempt : undefined,\n runId: c.req.query(\"run\") || undefined,\n sortKey: measurementSortKey(c.req.query(\"sort\")),\n direction,\n limit: positiveInteger(c.req.query(\"limit\"), 50, 100),\n offset: nonNegativeInteger(c.req.query(\"offset\"), 0, 10_000_000),\n });\n\n c.header(\"Cache-Control\", \"no-store\");\n c.header(\"X-Content-Type-Options\", \"nosniff\");\n return c.json({ measurements });\n});\n\napp.get(\"/source\", (c) => c.redirect(parseVal().links.self.val));\n\napp.onError((err, c) => {\n console.error(\n \"Benchmark request failed:\",\n err instanceof Error ? err.message : String(err),\n );\n c.header(\"Cache-Control\", \"no-store\");\n c.header(\"X-Content-Type-Options\", \"nosniff\");\n return c.req.path.startsWith(\"/api/\")\n ? c.json({ error: \"The request could not be completed.\" }, 500)\n : c.text(\"The page could not be loaded.\", 500);\n});\n\nexport default app.fetch;\n","label","deepseek-v41-flash",[162],"02af6a2e-1a38-455f-aeee-71b6df443e86","README.md","01a08da5-899f-73ab-baf1-ef75ca3ea3d7",7,["D",1789082872372],"# DeepSeek V4.1 Flash Gateway Benchmark\n\nA public, read-only Val Town page that compares providers serving\n`deepseek/deepseek-v4.1-flash` through Vercel AI Gateway.\n\n## How results work\n\nEvery provider attempt is one row in the val-scoped\n`benchmark_measurements` SQLite table. The default provider comparison pools\nmeasurements from every completed run for the model. Each run keeps its\nbenchmark version in SQLite so methodology changes remain auditable without\nremoving older completed measurements from the aggregate.\n\nBefore each run, the worker fetches Vercel's live endpoint catalog. It\nbenchmarks every returned provider and stores a normalized provider snapshot.\nProviders newly added by Vercel are included automatically. An interrupted run\nresumes against the exact provider set it started with.\n\nEach provider receives the same fixed multi-turn message array four times.\nAttempts for one provider run sequentially to preserve the cache warmup signal,\nwhile different providers run concurrently. Gateway automatic caching is\nenabled and each request is restricted to the provider being measured.\n\nThe sortable provider table reports:\n\n- Reliability across all attempts.\n- TTFT p50 and p95 across successful responses.\n- p50 and p5 observed output tokens per second after first text.\n- Cache hits among eligible repeated successful responses.\n- Observed cost per million input plus output tokens.\n\nTokens/sec is output tokens divided by `duration - TTFT`. It measures how\nquickly text reaches this client after the first text delta. Providers that\nbuffer output into a few large chunks can therefore appear unusually fast.\np50 is the median, TTFT p95 is the slow latency tail, and tokens/sec p5 is the\nslow throughput tail.\n\n$/MTok is total observed cost divided by total input plus output tokens,\nmultiplied by one million. Cached tokens remain part of input tokens, while\ntheir discount is reflected in observed cost.\n\nThe measurements page exposes Success and Error. Success is based on the\ncomplete consumed text stream, independent of the bounded private response\npreview. Error rows show a short, redacted message; timeouts include the\nconfigured 75 second limit. Raw response text and provider metadata remain\nprivate.\n\n## Schedule and recovery\n\n`benchmark/scheduled.ts` runs every 15 minutes in UTC. Requests have a\n75 second timeout and automatic SDK retries are disabled.\n\nA renewable SQLite lease prevents scheduled and manual starts from\noverlapping. The lease can be reclaimed after five minutes without a renewal,\nso an interrupted invocation does not block several later intervals.\nIndividual attempts are checkpointed, and a later invocation resumes the same\nbenchmark version without duplicating stored measurements.\n\nTo run manually, edit `benchmark/config.ts` if needed, open\n`benchmark/scheduled.ts` in Val Town, and click **Run**. Public visitors cannot\nstart model requests from the HTTP app.\n\n## Access\n\nThe Gateway credential is stored as the encrypted\n`VERCEL_AI_GATEWAY_API_KEY` environment variable. The HTTP app exposes only\nread-only routes.\n\n## Structure\n\n- `benchmark/prompt.ts` — fixed messages and benchmark version.\n- `benchmark/attempt.ts` — one streamed AI SDK request.\n- `benchmark/measurement.ts` — pure observation-to-row conversion.\n- `benchmark/endpoints.ts` — live Gateway provider discovery.\n- `benchmark/lease.ts` — renewable execution lease.\n- `benchmark/execute.ts` — benchmark orchestration.\n- `db/` — val-scoped SQLite schema, writes, provider snapshots, and aggregation.\n- `frontend/` — React interface styled with Twind.\n- `shared/` — shared benchmark and result types.\n",[173],"788d1e10-c0f1-4426-86dd-cda74350697d","AGENTS.md","81dba77c-ad6f-11f1-bc49-1607ee4eb77e",8,["D",1789082985670],"# deepseek-v41-flash-benchmark\n\nGateway provider benchmark for `deepseek/deepseek-v4.1-flash` through Vercel AI\nGateway. Remixed from `valdottown/glm53-flash-benchmark` with the model id and\nUI copy switched to DeepSeek. All providers, discovery, and metrics logic are\nidentical to the upstream.\n\n## File map\n\n| Path | Type | Purpose |\n|------|------|---------|\n| `index.ts` | http | Hono read-only app (`/`, `/measurements`, `/api/results`, `/api/measurements`) |\n| `benchmark/config.ts` | script | **Model + run config** — change `modelId` here |\n| `benchmark/prompt.ts` | script | Fixed multi-turn messages, `BENCHMARK_VERSION` |\n| `benchmark/attempt.ts` | script | One streamed AI SDK request |\n| `benchmark/measurement.ts` | script | Pure observation-to-row conversion |\n| `benchmark/endpoints.ts` | script | Live Gateway provider discovery |\n| `benchmark/lease.ts` | script | Renewable execution lease |\n| `benchmark/execute.ts` | script | Benchmark orchestration |\n| `benchmark/scheduled.ts` | interval | Cron entrypoint — runs every 15 min UTC |\n| `db/schema.ts` | script | SQLite schema + `sqlite` export |\n| `db/queries.ts` | script | Run/measurement writes and lookups |\n| `db/results.ts` | script | Aggregate provider comparison query |\n| `db/measurements.ts` | script | Paginated measurements listing |\n| `frontend/root.tsx` | script | HTML shell + title/meta |\n| `frontend/index.tsx` | script | Client entrypoint, page switch |\n| `frontend/components/App.tsx` | script | Model header + provider table view |\n| `frontend/components/ProviderTable.tsx` | script | Sortable provider table |\n| `frontend/components/MeasurementsPage.tsx` | script | Measurements page |\n| `frontend/components/Sparkline.tsx` | script | History sparklines |\n| `frontend/components/InfoPopover.tsx` | script | Column explainers |\n| `frontend/{providers,history,format}.ts` | script | Provider names, hourly series, formatters |\n| `frontend/favicon.svg` | file | Favicon (still the GLM-era glyph) |\n| `shared/types.ts` | script | Shared benchmark/result types |\n| `README.md` | file | Overview + schedule docs |\n\n## Credential\n\n`VERCEL_AI_GATEWAY_API_KEY` (scoped env var). HTTP app is read-only.\n",[180],"46eb97b6-37cb-403b-8d01-53b8d6528ad3","shared","01a08da5-899f-73ab-baf1-e8049921b5fd",["D",1789082831202],"directory",[187],"88251fa9-177a-4d3b-b712-fbb24a0ddaf6","frontend","01a08da5-899f-73ab-baf1-e025e2c5cf8b",["D",1789082831202],[193],"e6dde3be-d02c-487f-9b40-5146d7076871","db","01a08da5-899f-73ab-baf1-a881a20f339e",["D",1789082831202],[198],"7db55bd3-dcde-471f-a4e0-24c4c57a7efe","benchmark","01a08da5-899f-73ab-baf1-9427bec3af29",["D",1789082831202],[203],"pulls","branches","forks","envVars",{"_214":164,"_216":70},"connections",{"_214":70,"_215":70},"count","unfilledCount","emptyCount",{"_50":202,"_38":203,"_36":204,"_163":-5,"_48":49,"_47":164,"_165":66,"_45":463,"_13":70,"_40":190,"_42":-5,"_30":31,"_169":-5,"_252":464,"_33":465},{"_50":455,"_38":456,"_36":457,"_163":204,"_48":49,"_47":458,"_165":31,"_45":459,"_13":266,"_40":41,"_42":460,"_30":31,"_169":-5,"_252":461,"_33":462},{"_50":448,"_38":449,"_36":450,"_163":204,"_48":49,"_47":266,"_165":31,"_45":451,"_13":164,"_40":41,"_42":452,"_30":31,"_169":-5,"_252":453,"_33":454},{"_50":441,"_38":442,"_36":443,"_163":204,"_48":49,"_47":164,"_165":66,"_45":444,"_13":70,"_40":41,"_42":445,"_30":31,"_169":-5,"_252":446,"_33":447},{"_50":437,"_38":39,"_36":37,"_163":204,"_48":49,"_47":14,"_165":66,"_45":438,"_13":266,"_40":41,"_42":44,"_30":31,"_169":-5,"_252":439,"_33":440},{"_50":430,"_38":431,"_36":432,"_163":204,"_48":49,"_47":164,"_165":66,"_45":433,"_13":70,"_40":41,"_42":434,"_30":31,"_169":-5,"_252":435,"_33":436},{"_50":423,"_38":424,"_36":425,"_163":204,"_48":49,"_47":164,"_165":31,"_45":426,"_13":70,"_40":41,"_42":427,"_30":31,"_169":-5,"_252":428,"_33":429},{"_50":416,"_38":417,"_36":418,"_163":204,"_48":49,"_47":164,"_165":31,"_45":419,"_13":70,"_40":41,"_42":420,"_30":31,"_169":-5,"_252":421,"_33":422},{"_50":408,"_38":409,"_36":410,"_163":204,"_48":49,"_47":164,"_165":66,"_45":411,"_13":70,"_40":412,"_42":413,"_30":31,"_169":-5,"_252":414,"_33":415},{"_50":400,"_38":401,"_36":402,"_163":204,"_48":49,"_47":403,"_165":31,"_45":404,"_13":70,"_40":41,"_42":405,"_30":31,"_169":-5,"_252":406,"_33":407},{"_50":197,"_38":198,"_36":199,"_163":-5,"_48":49,"_47":164,"_165":66,"_45":397,"_13":70,"_40":190,"_42":-5,"_30":31,"_169":-5,"_252":398,"_33":399},{"_50":390,"_38":391,"_36":392,"_163":199,"_48":49,"_47":164,"_165":66,"_45":393,"_13":70,"_40":41,"_42":394,"_30":31,"_169":-5,"_252":395,"_33":396},{"_50":383,"_38":384,"_36":385,"_163":199,"_48":49,"_47":164,"_165":66,"_45":386,"_13":70,"_40":41,"_42":387,"_30":31,"_169":-5,"_252":388,"_33":389},{"_50":376,"_38":377,"_36":378,"_163":199,"_48":49,"_47":164,"_165":31,"_45":379,"_13":70,"_40":41,"_42":380,"_30":31,"_169":-5,"_252":381,"_33":382},{"_50":369,"_38":370,"_36":371,"_163":199,"_48":49,"_47":164,"_165":66,"_45":372,"_13":70,"_40":41,"_42":373,"_30":31,"_169":-5,"_252":374,"_33":375},{"_50":192,"_38":193,"_36":194,"_163":-5,"_48":49,"_47":164,"_165":66,"_45":366,"_13":70,"_40":190,"_42":-5,"_30":31,"_169":-5,"_252":367,"_33":368},{"_50":362,"_38":324,"_36":319,"_163":194,"_48":49,"_47":164,"_165":66,"_45":363,"_13":70,"_40":190,"_42":-5,"_30":31,"_169":-5,"_252":364,"_33":365},{"_50":353,"_38":354,"_36":355,"_163":319,"_48":49,"_47":356,"_165":31,"_45":357,"_13":358,"_40":41,"_42":359,"_30":31,"_169":-5,"_252":360,"_33":361},{"_50":346,"_38":347,"_36":348,"_163":319,"_48":49,"_47":164,"_165":66,"_45":349,"_13":70,"_40":41,"_42":350,"_30":31,"_169":-5,"_252":351,"_33":352},{"_50":339,"_38":340,"_36":341,"_163":319,"_48":49,"_47":164,"_165":66,"_45":342,"_13":70,"_40":41,"_42":343,"_30":31,"_169":-5,"_252":344,"_33":345},{"_50":332,"_38":333,"_36":334,"_163":319,"_48":49,"_47":164,"_165":66,"_45":335,"_13":70,"_40":41,"_42":336,"_30":31,"_169":-5,"_252":337,"_33":338},{"_50":325,"_38":326,"_36":327,"_163":319,"_48":49,"_47":164,"_165":31,"_45":328,"_13":70,"_40":41,"_42":329,"_30":31,"_169":-5,"_252":330,"_33":331},{"_50":316,"_38":317,"_36":318,"_163":319,"_48":49,"_47":164,"_165":66,"_45":320,"_13":70,"_40":41,"_42":321,"_30":31,"_169":-5,"_252":322,"_33":323},{"_50":309,"_38":310,"_36":311,"_163":194,"_48":49,"_47":164,"_165":66,"_45":312,"_13":70,"_40":26,"_42":313,"_30":31,"_169":-5,"_252":314,"_33":315},{"_50":302,"_38":303,"_36":304,"_163":194,"_48":49,"_47":164,"_165":66,"_45":305,"_13":70,"_40":41,"_42":306,"_30":31,"_169":-5,"_252":307,"_33":308},{"_50":295,"_38":296,"_36":297,"_163":194,"_48":49,"_47":164,"_165":66,"_45":298,"_13":70,"_40":41,"_42":299,"_30":31,"_169":-5,"_252":300,"_33":301},{"_50":288,"_38":289,"_36":290,"_163":194,"_48":49,"_47":164,"_165":66,"_45":291,"_13":70,"_40":41,"_42":292,"_30":31,"_169":-5,"_252":293,"_33":294},{"_50":281,"_38":282,"_36":283,"_163":194,"_48":49,"_47":164,"_165":31,"_45":284,"_13":70,"_40":41,"_42":285,"_30":31,"_169":-5,"_252":286,"_33":287},{"_50":273,"_38":274,"_36":275,"_163":194,"_48":49,"_47":276,"_165":66,"_45":277,"_13":164,"_40":41,"_42":278,"_30":31,"_169":-5,"_252":279,"_33":280},{"_50":186,"_38":187,"_36":188,"_163":-5,"_48":49,"_47":164,"_165":66,"_45":270,"_13":70,"_40":190,"_42":-5,"_30":31,"_169":-5,"_252":271,"_33":272},{"_50":261,"_38":262,"_36":263,"_163":188,"_48":49,"_47":264,"_165":31,"_45":265,"_13":266,"_40":41,"_42":267,"_30":31,"_169":-5,"_252":268,"_33":269},{"_50":179,"_38":180,"_36":181,"_163":-5,"_48":49,"_47":182,"_165":31,"_45":258,"_13":70,"_40":26,"_42":184,"_30":31,"_169":-5,"_252":259,"_33":260},{"_50":172,"_38":173,"_36":174,"_163":-5,"_48":49,"_47":175,"_165":31,"_45":255,"_13":164,"_40":26,"_42":177,"_30":31,"_169":-5,"_252":256,"_33":257},{"_50":161,"_38":162,"_36":64,"_163":-5,"_48":49,"_47":164,"_165":66,"_45":251,"_13":70,"_40":167,"_42":168,"_30":31,"_169":170,"_252":253,"_33":254},["D",1789082831202],"children",[],[162],["D",1789082872372],[],[173],["D",1789082985670],[],[180],"25f218dd-cb4f-4bea-8f58-371e8cec9f57","types.ts","01a08da5-899f-73ab-baf1-e4b182203ecc",10,["D",1789484573494],2,"export type BenchmarkStatus = \"running\" | \"completed\" | \"failed\";\nexport type MeasurementStatus = \"ok\" | \"error\";\n\nexport interface BenchmarkConfig {\n benchmarkVersion: string;\n label: string;\n modelId: string;\n runsPerProvider: number;\n maxOutputTokens: number;\n providerConcurrency: number;\n requestTimeoutMs: number;\n providers?: string[];\n}\n\n/**\n * \"provider\" rows pin one provider with `only: [name]`. \"strategy\" rows are\n * synthetic endpoints that exercise a way of calling the gateway (default\n * autorouting, `sort`, `order`, handpicked `only` sets…).\n */\nexport type EndpointKind = \"provider\" | \"strategy\";\n\nexport interface GatewayEndpoint {\n provider_name: string;\n kind?: EndpointKind;\n description?: string;\n gateway_options?: Record;\n name?: string;\n has_zdr?: boolean;\n has_no_training?: boolean;\n supports_implicit_caching?: boolean;\n pricing?: Record;\n latency_last_1h?: { p50?: number; p95?: number };\n throughput_last_1h?: { p50?: number; p95?: number };\n tags?: string[];\n [key: string]: unknown;\n}\n\nexport interface BenchmarkRunSummary {\n id: string;\n label: string;\n modelId: string;\n status: BenchmarkStatus;\n startedAt: string;\n completedAt: string | null;\n runsPerProvider: number;\n maxOutputTokens: number;\n providerCount: number;\n successfulMeasurements: number;\n failedMeasurements: number;\n error: string | null;\n}\n\nexport type StoredMeasurementOutcome = \"text\" | \"no_text\" | \"error\";\nexport type MeasurementErrorCode =\n | \"no_text\"\n | \"timeout\"\n | \"provider_error\"\n | null;\n\nexport interface BenchmarkMeasurement {\n runId: string;\n provider: string;\n attempt: number;\n status: MeasurementStatus;\n outcome: StoredMeasurementOutcome;\n errorCode: MeasurementErrorCode;\n startedAt: string;\n completedAt: string;\n ttftMs: number | null;\n durationMs: number | null;\n inputTokens: number | null;\n outputTokens: number | null;\n cacheReadTokens: number | null;\n cacheWriteTokens: number | null;\n costUsd: number | null;\n tokensPerSecond: number | null;\n responsePreview: string | null;\n error: string | null;\n providerMetadata: Record | null;\n providerInfo: GatewayEndpoint;\n}\n\nexport type MeasurementPublicStatus = \"success\" | \"error\";\n\nexport type MeasurementSortKey =\n | \"runStartedAt\"\n | \"provider\"\n | \"attempt\"\n | \"status\"\n | \"startedAt\"\n | \"ttftMs\"\n | \"durationMs\"\n | \"tokensPerSecond\"\n | \"inputTokens\"\n | \"outputTokens\"\n | \"cacheReadTokens\"\n | \"cacheWriteTokens\"\n | \"costUsd\"\n | \"message\";\n\nexport type MeasurementSortDirection = \"asc\" | \"desc\";\n\nexport interface MeasurementListItem {\n runId: string;\n runStartedAt: string;\n runStatus: BenchmarkStatus;\n provider: string;\n attempt: number;\n status: MeasurementPublicStatus;\n startedAt: string;\n completedAt: string;\n ttftMs: number | null;\n durationMs: number | null;\n inputTokens: number | null;\n outputTokens: number | null;\n cacheReadTokens: number | null;\n cacheWriteTokens: number | null;\n costUsd: number | null;\n tokensPerSecond: number | null;\n error: string | null;\n}\n\nexport interface MeasurementRunOption {\n id: string;\n startedAt: string;\n status: BenchmarkStatus;\n measurementCount: number;\n}\n\nexport interface MeasurementListResponse {\n items: MeasurementListItem[];\n total: number;\n limit: number;\n offset: number;\n providers: string[];\n attempts: number[];\n runs: MeasurementRunOption[];\n}\n\nexport interface BenchmarkRunDetail extends BenchmarkRunSummary {\n config: BenchmarkConfig;\n endpointSnapshot: GatewayEndpoint[];\n measurements: BenchmarkMeasurement[];\n}\n\nexport interface ProviderMetricPoint {\n runId: string;\n runStartedAt: string;\n successRate: number;\n attempts: number;\n successes: number;\n p50TtftMs: number | null;\n p95TtftMs: number | null;\n p50TokensPerSecond: number | null;\n p05TokensPerSecond: number | null;\n cacheHitRate: number | null;\n effectiveCostPerMillionTokensUsd: number | null;\n}\n\nexport interface ProviderResult {\n provider: string;\n kind: EndpointKind;\n /** Zero data retention per the Vercel endpoint catalogue (null = unknown). */\n hasZdr: boolean | null;\n /** Provider does not train on prompts per the catalogue (null = unknown). */\n hasNoTraining: boolean | null;\n /** providerOptions.gateway used by a strategy row; null for provider rows. */\n gatewayOptions: Record | null;\n /** Successful attempts by the provider the gateway finally routed to. */\n routedProviders: Record;\n /** Successful attempts where the gateway retried after a provider failure. */\n recoveredAttempts: number;\n runCount: number;\n attempts: number;\n successes: number;\n failures: number;\n successRate: number;\n p50TtftMs: number | null;\n p95TtftMs: number | null;\n p50DurationMs: number | null;\n p95DurationMs: number | null;\n p50TokensPerSecond: number | null;\n p05TokensPerSecond: number | null;\n cacheEligibleAttempts: number;\n cacheHits: number;\n cacheHitRate: number | null;\n effectiveCostPerMillionTokensUsd: number | null;\n history: ProviderMetricPoint[];\n}\n\nexport interface BenchmarkResults {\n modelId: string;\n runCount: number;\n attemptCount: number;\n successfulAttempts: number;\n failedAttempts: number;\n firstRunAt: string | null;\n lastCompletedAt: string | null;\n providers: ProviderResult[];\n}\n",[],[187,262],["D",1789082831202],[247],[187],"70e5d894-3ea5-4a94-9391-a86dad9e083f","root.tsx","01a08da5-899f-73ab-baf1-df2fd8ed278d",6,["D",1789082872323],"/** @jsxImportSource npm:hono@4/jsx */\nimport { raw } from \"npm:hono@4/html\";\nimport { immutableFileUrl } from \"https://esm.town/v/std/utils/index.ts\";\n\nexport function Root() {\n return (\n <>\n {raw(\"\")}\n \n \n \n \n \n \n DeepSeek V4.1 Flash Provider Benchmark\n

Diff · v15