Machine-readable LLM API pricing for 156 current models across 28 providers (OpenAI, Anthropic, Google, Mistral, xAI, DeepSeek, Meta, Groq, Together, …), maintained by modelpricewatch.com. Every price links to the provider's official pricing page; prices are cross-checked daily against an independent reference and top models are re-read weekly from the provider pages with dated receipts.
Last sync: 2026-08-05. This repo is updated automatically after each refresh cycle.
| File | What it is |
|---|---|
models.json |
Every tracked model: input/output/cached $ per 1M tokens, context window, modality, category, status, source URL |
providers.json |
Provider metadata + official pricing-page URLs |
price-history.json / price-history.csv |
Per-model price timeline (6975 dated points; see caveats) |
llm-price-index.csv |
The LLM Price Index — one number for the blended cost of frontier intelligence, daily series |
CHANGELOG.md |
Human-readable launches & price changes |
mcp/ |
MCP server: query these prices live from Claude/Cursor — npx -y @modelpricewatch/mcp or HTTP https://modelpricewatch.com/mcp |
The same data, always current, CORS-open:
https://modelpricewatch.com/api/v1/models.json
https://modelpricewatch.com/api/v1/price-history.json
Docs: https://modelpricewatch.com/api/ · Price-change feed (Atom): https://modelpricewatch.com/news/feed.xml
A fixed, equal-weighted basket of frontier flagship models, blended 3:1 input:output, re-verified daily — a single citeable number for "what does a million tokens of frontier intelligence cost?" Methodology: https://modelpricewatch.com/price-index/methodology/
- History points are labelled by
source: our ownverified/snapshot captures vslitellm-archive(reconstructed from the open-source LiteLLM repo's git history — marked, never mixed silently). - A deeper 2024→2026 history backfill is in progress; some models currently have short timelines. Known baseline discontinuities are under audit.
Data: CC-BY-4.0 — free for any use, with attribution to modelpricewatch.com. MCP bridge code: MIT.