--- title: AI Pricing API description: Public JSON API documentation for current AI model provider pricing. doc_version: "1.0" last_updated: "2026-05-07" --- # AI Pricing API Read-only HTTP API for current pricing across the providers we track. No authentication required. Responses are JSON. Rate-limited per IP per endpoint group (60 req / 60 s on `/v1/prices/*`, 120 req / 60 s on the rest). **Base URL:** `https://ai-pricing.fyi` **Get this doc as Markdown:** `curl -H 'Accept: text/markdown' https://ai-pricing.fyi/api` **OpenAPI schema:** `https://ai-pricing.fyi/openapi.json` --- ## Contents - [Providers](#providers) — who hosts models - [Models](#models) — canonical catalog of every model we have data for - [Prices](#prices) — current per-token / per-call rates - [Common queries](#common-queries) — recipes for typical questions - [Sitemap](#sitemap) — related machine-readable resources --- ## Providers A provider is an inference platform we scrape (Anthropic, OpenAI, Cloudflare Workers AI, Fireworks AI, Together AI). The same model can be hosted by multiple providers — each provider's offer points at the same canonical model row, so prices stack on the model detail page. ### `GET /v1/providers` List every provider we track. Pass `active=true` to exclude any future deactivated ones. | Param | Type | Description | |---|---|---| | `active` | boolean | `true` \| `false`. Omit to return all. | | `limit` | integer | Default 50. | | `offset` | integer | Default 0. | ```bash curl https://ai-pricing.fyi/v1/providers ``` ```json { "data": [ { "id": 2, "slug": "anthropic", "name": "Anthropic", "website_url": "https://anthropic.com", "docs_url": "https://docs.anthropic.com", "pricing_url": "https://platform.claude.com/docs/en/about-claude/pricing", "icon_url": "https://www.google.com/s2/favicons?domain=anthropic.com&sz=64", "active": true, "notes": null, "created_at": "2026-04-22 19:08:49", "updated_at": "2026-04-22 19:08:49" } ], "limit": 50, "offset": 0 } ``` ### `GET /v1/providers/:slug` Detail for one provider, by its slug (e.g. `anthropic`, `openai`, `cloudflare`, `fireworks`, `together`). ```bash curl https://ai-pricing.fyi/v1/providers/anthropic ``` --- ## Models One row per canonical model in our catalog. `vendor` is the canonical model creator/vendor (often the original creator), not necessarily the serving provider. `family` is a coarse grouping like `opus`, `sonnet`, `gpt`, `kimi`. `status` is one of `active`, `deprecated` (provider has announced a sunset date), or `retired` (no longer offered). ### `GET /v1/models` List models in the catalog. Combine `vendor` + `family` to drill down by canonical creator, or use `provider` + `deployment_mode` to find what a serving provider offers. | Param | Type | Description | |---|---|---| | `vendor` | string | Canonical model creator/vendor slug, not the serving provider. Example: `anthropic` returns Anthropic-created models, including models served through other providers. | | `provider` | string | Serving provider slug. Example: `cloudflare` returns models Cloudflare can serve, including proxied AI Gateway catalog entries. | | `deployment_mode` | string | Filter by serving mode such as `hosted`, `proxied`, or `shared_serverless`. | | `family` | string | `opus`, `sonnet`, `haiku`, `gpt`, `kimi`, `glm`, … | | `open_weights` | boolean | `true` \| `false` | | `model_type` | string | Filter by canonical model type. Comma-list for OR. Values: `chat`, `reasoning`, `code`, `embedding`, `multimodal_chat`, `image_generation`, `audio_transcription`, `audio_generation`, `video_generation`, `ocr`, `moderation`. | | `input_modality` | string | Filter to models accepting this modality. Repeat to AND. Values: `text`, `image`, `audio`, `video`. | | `output_modality` | string | Filter to models producing this modality. Repeat to AND. Values: `text`, `image`, `audio`, `video`, `vector`. | | `capability` | string | Filter to models with this capability flag. Repeat to AND. Values: `tool_use`, `json_mode`, `thinking_mode`, `web_search`, `code_execution`, `file_attachments`. | | `model_group` | string | Filter to canonical models whose `model_group_key` equals the given value (e.g. `meta/llama-3.3-70b-instruct`). Use this to find every per-provider variant of the same underlying model. | | `modality` | string | **(DEPRECATED — alias for `input_modality`)** When set on the model row. | | `limit` | integer | Default 50. Pass 1000 to get the full catalog. | | `offset` | integer | Default 0. | Response rows include a `model_group_key` field (nullable string). When non-null, it groups together canonical rows that represent the same underlying model hosted under different per-provider slugs (e.g. Groq's `llama-3.3-70b-versatile` and Cloudflare's `llama-3.3-70b-instruct-fp8-fast` both share the group key `meta/llama-3.3-70b-instruct`). Rows also include `max_output_tokens` (nullable integer) when the provider publishes a separate max-output limit distinct from `context_window`. Only providers whose pricing / detail pages expose this number populate it; expect `null` on older or scraper-less providers. ```bash curl "https://ai-pricing.fyi/v1/models?vendor=anthropic&limit=2" curl "https://ai-pricing.fyi/v1/models?provider=cloudflare&deployment_mode=proxied&limit=20" ``` ```json { "data": [ { "id": 52, "canonical_slug": "claude-haiku-3", "vendor": "anthropic", "family": "haiku", "model_name": "claude-haiku-3", "display_name": "Claude Haiku 3", "status": "active", "open_weights": false, "context_window": null, "max_output_tokens": null, "params_billions": null, "model_type": "chat", "input_modalities": ["text"], "output_modalities": ["text"], "capabilities": ["tool_use", "json_mode"], "model_group_key": "anthropic/claude-haiku-3", "created_at": "2026-04-23 01:15:25", "updated_at": "2026-04-23 01:15:25" } ], "limit": 2, "offset": 0 } ``` ### `GET /v1/models/:key` Detail for one model. Includes any aliases that resolve to the same canonical row. The `:key` path segment accepts either a canonical slug (`llama-3.3-70b-versatile`) or a model group key (`meta/llama-3.3-70b-instruct`): - Given a **canonical slug** whose row has a non-null `model_group_key`, the endpoint returns `301 Moved Permanently` with `Location: /v1/models/`. Clients that follow redirects land transparently on the group view. - Given a **group key**, returns the representative row (the first canonical model alphabetically that shares that group key) plus aggregate capability data across the group. - `404` when the value matches neither a canonical slug nor a group key. ```bash curl https://ai-pricing.fyi/v1/models/claude-opus-4.7 curl -L https://ai-pricing.fyi/v1/models/meta/llama-3.3-70b-instruct curl "https://ai-pricing.fyi/v1/models/claude-opus-4.7/offers" ``` Cloudflare AI Gateway proxied offers are availability rows. They appear in `/v1/models/:canonicalSlug/offers` with `pricing_status: "proxied"` and do not appear in `/v1/prices/current`. --- ## Prices Current scraped prices, one row per (offer × dimension). One model can produce many rows: one per provider that hosts it, one per tier (standard / batch), one per metric (`input` / `cached_input` / `cache_write_5m` / `cache_write_1h` / `output`). Filter aggressively. ### `GET /v1/prices/current` The main endpoint. Returns the latest known numeric price for every active priced offer matching the filters. | Param | Type | Description | |---|---|---| | `provider` | string | Provider slug (`anthropic`, `openai`, `cloudflare`, `fireworks`, `together`). | | `model` | string | Canonical slug (e.g. `claude-opus-4.7`, `gpt-5.4`, `kimi-k2.5`). Returns rows from all providers hosting this model. | | `family` | string | `opus`, `sonnet`, `gpt`, `kimi`, … | | `model_type` | string | Filter by canonical model type. Comma-list for OR. Values: `chat`, `reasoning`, `code`, `embedding`, `multimodal_chat`, `image_generation`, `audio_transcription`, `audio_generation`, `video_generation`, `ocr`, `moderation`. | | `input_modality` | string | Filter to models accepting this modality. Repeat to AND. Values: `text`, `image`, `audio`, `video`. | | `output_modality` | string | Filter to models producing this modality. Repeat to AND. Values: `text`, `image`, `audio`, `video`, `vector`. | | `capability` | string | Filter to models with this capability flag. Repeat to AND. Values: `tool_use`, `json_mode`, `thinking_mode`, `web_search`, `code_execution`, `file_attachments`. | | `model_group` | string | Filter price rows whose canonical model has the given `model_group_key` (e.g. `meta/llama-3.3-70b-instruct`). This is the canonical way to compare prices for the same underlying model across providers. | | `metric` | string | `input_token`, `cached_input_token`, `output_token`, `cache_write_5m_token`, `cache_write_1h_token`. | | `billing_basis` | string | `per_token`, `per_usage`. | | `billing_tier` | string | `standard`, `batch`, `flex`, `priority`, `turbo`. Matches the suffix on each row's `provider_offer_key`. | | `region` | string | AWS/GCP region for region-priced offers. | | `limit` | integer | Default 100. Max useful ~2000. | | `offset` | integer | Default 0. | ```bash curl "https://ai-pricing.fyi/v1/prices/current?model=claude-opus-4.7&limit=1" ``` ```json { "data": [ { "offer_id": 176, "provider_slug": "anthropic", "provider_name": "Anthropic", "canonical_slug": "claude-opus-4.7", "display_name": "Claude Opus 4.7", "model_name": "claude-opus-4.7", "family": "opus", "vendor": "anthropic", "open_weights": false, "model_group_key": null, "provider_offer_key": "anthropic_claude-opus-4.7_standard", "offer_category": "model_api", "deployment_mode": "shared_serverless", "billing_basis": "per_token", "metric": "input_token", "unit": "per_1m_tokens", "currency": "USD", "batch_flag": 0, "price_numeric": 5, "latest_observed_at": "2026-04-24T00:14:27.018Z", "region": null } ] } ``` Response rows include a `model_group_key` field (nullable string) that matches the column on `/v1/models`. Use it together with the `model_group` query param to fetch every hosting provider's offer for the same underlying model: ```bash curl "https://ai-pricing.fyi/v1/prices/current?model_group=meta/llama-3.3-70b-instruct" ``` Returns all hosting providers' offers for Meta's Llama 3.3 70B Instruct. ### `GET /v1/prices/filters` Discover what values are available for the filter parameters above. Useful for building dropdowns. ```bash curl https://ai-pricing.fyi/v1/prices/filters ``` ```json { "providers": ["anthropic", "cloudflare", "fireworks", "openai", "together"], "families": ["opus", "sonnet", "haiku", "gpt", "kimi", "..."], "model_types": ["chat", "reasoning", "code", "embedding", "..."], "input_modalities": ["text", "image", "audio", "video"], "output_modalities": ["text", "image", "audio", "video", "vector"], "capabilities": ["tool_use", "json_mode", "thinking_mode", "web_search", "code_execution", "file_attachments"], "metrics": ["input_token", "cached_input_token", "output_token", "..."], "billing_basis": ["per_token"], "billing_tiers": ["standard", "batch", "flex", "priority", "turbo"], "comparable_classes": ["..."] } ``` --- ## Common queries ### All providers we have data for ```bash curl https://ai-pricing.fyi/v1/providers ``` ### All models for a single provider (active only) ```bash curl "https://ai-pricing.fyi/v1/models?provider=fireworks&limit=1000" ``` Filter the response by `status === "active"` if you want to hide deprecated/retired entries. ### All current prices for one model across every provider hosting it ```bash curl "https://ai-pricing.fyi/v1/prices/current?model=gpt-oss-120b&limit=100" ``` For shared open-weight models you'll get rows from multiple providers in one call — the `provider_slug` field tells you which is which. ### Cheapest input-token price for every Anthropic model right now ```bash curl "https://ai-pricing.fyi/v1/prices/current?provider=anthropic&metric=input_token&billing_tier=standard&limit=200" ``` ### Compare batch vs standard pricing for one model ```bash curl "https://ai-pricing.fyi/v1/prices/current?model=claude-opus-4.7&limit=20" ``` Group the response client-side by `provider_offer_key` — keys ending in `_standard` vs `_batch` are the two tiers. ## Sitemap See the full [sitemap](/sitemap.md) for all public pages and machine-readable resources.