Price Per Token

About Price Per Token

Price Per Token is a continuously updated reference for what it actually costs to call a large language model. It tracks more than 300 models across the major API providers — OpenAI, Anthropic, Google, Mistral, DeepSeek, Moonshot, Meta, Amazon, Cohere, Qwen, IBM and a long tail of smaller hosts — and puts input price, output price, cached-input price, and context window in one sortable grid so you can see the real spread between a frontier model and a small one at a glance. Alongside price it carries benchmark columns (coding, MMLU, GPQA) so a cost comparison does not have to be made blind to capability, and capability filters for tool use, vision, reasoning, web search, PDF input, prompt caching, audio input, and open-source licensing let you narrow the grid to only the models that can do the job you have. A new-releases rail tracks models as they ship, which in practice is the fastest way to notice that a provider has quietly cut prices or launched a cheaper tier. There is also an MCP server, so a coding agent can query live pricing and benchmark data mid-task rather than reasoning from a stale training snapshot, plus a weekly newsletter covering price changes and new releases.

Key Features

Tags