Price Per Token

Price Per Token is a reference site for anyone trying to figure out what an LLM API call will actually cost. It tracks per-token input/output pricing for 300+ models from OpenAI, Anthropic, Google, Meta, Mistral, DeepSeek and others, and surfaces that data through a pricing calculator, a token counter, side-by-side model comparisons, and benchmark leaderboards (MMLU-Pro, GPQA, LiveCodeBench, AIME). The dataset is refreshed as providers announce changes, which is the entire reason a tool like this exists.

Where it gets interesting is the Price Per Token MCP server: a free, no-auth MCP endpoint at api.pricepertoken.com/mcp/mcp that lets agents inside Claude Code, Cursor and Windsurf query live pricing and benchmark data at runtime. That turns a static price chart into something an autonomous coding agent can actually consult before picking a model. It's pitched at developers, engineering leads and AI ops folks who need to budget token spend or pick a cost/quality sweet spot, not at end users writing prose.

The site itself is free and ad-supported; there is no paid tier, no signup, and no API key wall. The main caveats are the usual ones for any aggregated pricing dataset: list prices don't reflect committed-use discounts, batch APIs or self-hosted economics, and benchmark scores are only as honest as the labs reporting them.