AI model pricing data,
free for everyone
A community-friendly JSON API exposing the same pricing data that powers the calculator. LiteLLM-verified, reconciled daily, CDN-cached. Build cost estimators, comparison tools, internal dashboards — no signup required.
Just fetch it.
No API key, no rate limits for normal use, no signup. Hit any endpoint with a GET request.
# Get all models
curl https://aicostcalc.net/api/v1/models
# Get a single model
curl https://aicostcalc.net/api/v1/models/gpt-5-5
# Get just pricing (lighter payload)
curl https://aicostcalc.net/api/v1/pricing?provider=anthropicAll responses are JSON, all CORS-enabled, all aggressively CDN-cached (1 hour browser, 24 hour edge).
Four endpoints. Zero ceremony.
/api/v1/modelsAll 50 models with full data (pricing, limits, capabilities, lifecycle). New models appear the day they reach the LiteLLM registry.
GET /api/v1/models
GET /api/v1/models?provider=anthropic
GET /api/v1/models?category=flagship
GET /api/v1/models?capability=caching
GET /api/v1/models?status=activeprovider / category / capability / status. Filters compose with AND./api/v1/models/{id}Full data for a single model. Accepts the dashed id or the dotted form of the model name; when a non-canonical form is used, _meta.resolvedFrom records what you asked for. Returns 404 with availableIds if nothing matches.
GET /api/v1/models/gpt-6-luna
GET /api/v1/models/gpt-6.luna # same model
GET /api/v1/models/claude-opus-4-7/api/v1/pricingPrices and limits only, plus lifecycle (status, deprecatedAt, successorId) so you never route to a retired model. Same filters as /models.
GET /api/v1/pricing
GET /api/v1/pricing?provider=openai
GET /api/v1/pricing?status=active # exclude retired models/api/v1/changesNew models and price changes, newest first — the JSON form of the changelog. Each event carries before/after prices and the change in per-call cost. Our own data corrections are excluded.
GET /api/v1/changes
GET /api/v1/changes?since=2026-09-01
GET /api/v1/changes?kind=price-cut,price-rise
GET /api/v1/changes?provider=anthropic&limit=20What you get back
Full schema reference is in our TypeScript types file. The shape of a model entry:
{
"id": "gpt-5-5",
"name": "GPT-5.5",
"shortName": "GPT-5.5",
"provider": "OpenAI",
"providerId": "openai",
"category": "flagship", // flagship | small | reasoning | balanced
"useCase": ["general", "vision", "reasoning", "coding"],
"releaseDate": "2026-04-23", // null when unknown (v2.1) — see below
"status": "active", // active | deprecated | preview | legacy
"deprecatedAt": null,
"successorId": null,
"pricing": {
"currency": "USD",
"unit": "per_1m_tokens",
"input": 5.00,
"output": 30.00,
"cachedInput": 0.50, // null if not supported
"cacheWrite": null, // null if not supported
"batchInput": 2.50, // null if not supported
"batchOutput": 15.00,
"imagePerImage": null, // null until image pricing added (V1.1)
"reasoningOutput": null,
"fineTunedInput": null,
"fineTunedOutput": null
},
"limits": {
"contextWindow": 1050000,
"maxOutput": 128000,
"maxImagesPerRequest": null,
"knowledgeCutoff": "2026-02"
},
"tokenization": {
"encoder": "o200k_base", // tiktoken encoder, or "approximate"
"approximationRatio": { "english": 4, "chinese": 1.5, "code": 3 }
},
"supports": {
"vision": true, "audio": false, "tools": true, "streaming": true,
"caching": true, "batch": true, "fineTuning": false, "structuredOutput": true
},
"lastVerified": "2026-05-12",
"sources": [
{ "type": "community", "url": "...", "fetchedAt": "..." },
{ "type": "official", "url": "...", "fetchedAt": "..." }
],
"priceHistory": [
{ "date": "2026-04-23", "input": 5.00, "output": 30.00, "note": "Initial release" }
],
"i18n": {
"en": { "tagline": "...", "description": "..." },
"zh": { "tagline": "...", "description": "..." }
},
"draft": { // present only on auto-listed models (v2.1)
"generatedAt": "2026-09-29",
"note": "..."
}
}releaseDatecan benull. Models listed automatically from the registry have no known release date, and we don't invent one. Type it asstring | null.draftis an optional object on models published automatically the day they appeared. Prices and limits on a draft are as reliable as any other entry; only the prose description is awaiting review./api/v1/pricingnow includesstatus,deprecatedAtandsuccessorId, and supports all four filters.
No field was renamed or removed. The one change that can break a strictly typed client is releaseDate becoming nullable.
Where the numbers come from
- ✓Primary source: LiteLLM's public model registry — the de facto industry standard, used by millions of LangChain and LiteLLM installations. Errors get caught and PR'd by the community within hours.
- ✓Secondary source: Each provider's official pricing page (links in every model's
sourcesfield). - ✓Refresh cadence: Daily — prices are reconciled against the LiteLLM registry every morning (06:15 UTC) and corrections deploy automatically against both sources, with diff review before publication.
- ⚠Disclaimer: This is informational data. For final billing accuracy, always verify against the provider's own pricing page. Spotted an error? Open a GitHub issue — corrections typically deploy within 24 hours.
Some ideas
MIT
The data is free to use commercially, embed in products, redistribute. Attribution appreciated but not required. The same license applies to the source code.
None for normal use
Vercel CDN handles the load. Cached at edge for 24 hours, so most requests never even hit the origin. If you're planning 10K+ requests/hour, please say hi so we can size accordingly.