LLM Token Costs & Pricing
Maintained by Coolhand Labs — agents that improve your agents
Token costs — the price per input/output token each provider charges — are the core unit of LLM tokenomics. This page tracks live cost-per-token and deprecation dates across 100+ models so you can compare providers without digging through their docs. Prefer code to a table? The same data is available as a free, no-API-key-required JSON API.
Coolhand does more than track pricing.
We watch your production AI agents end-to-end — deprecated models, silent failures, drifting quality — and propose the fix as a pull request. Free to start.
Live Pricing & Deprecation Status
No LLM models available yet.
This data is part of Coolhand
Coolhand helps engineering teams collect human feedback on AI outputs and automatically improve their LLM prompts.
Frequently asked questions
Data is gathered automatically by Coolhand's Godfrey agent, which periodically fetches and parses pricing pages published by model providers. Every record is reviewed before publishing, but it's still at the mercy of providers changing prices without much notice. The Sources link on each row points to the original provider pricing page that record came from.
Pricing is gathered automatically by Coolhand's Godfrey agent, which periodically fetches and parses pricing pages published by model providers. Every change is reviewed before publishing, so this page reflects provider-published rates without the lag of a manual audit.
Prices are stored as cost-per-token in USD, but the table shows the standard per-million-token format by default since that's how most providers publish their rates. Use the "Show" toggle above the table to switch to raw per-token pricing.
Multiply each model's per-token cost by how many input/output tokens your workload actually consumes to get total spend — this page gives you the current per-token and per-million-token price for every model, so you only need to supply your own usage volume. Pull the same data programmatically via GET /api/v2/inference_apis to wire it into your own cost-estimation tooling.
Providers publish input and output rates separately, usually as dollars per million tokens. Cost-per-token is just that rate divided by 1,000,000 — this page stores the per-token figure and lets you toggle between the two formats with the "Show" control above the table.
Providers deprecate models on their own timeline, often with little notice — requests to a deprecated model can start failing or silently return degraded results. This page shows each model's deprecation date and notes as soon as a provider publishes them, so you can see it here before it shows up as an incident in production.
Coolhand watches your production AI agents and can flag deprecated or soon-to-be-deprecated models automatically, proposing the fix as a pull request instead of leaving you to notice after requests start failing.
Yes — this page and its data are free to use, no API key required. The same data is available as a JSON API at GET /api/v2/inference_apis (unauthenticated requests are limited to 1 per minute; requests with a free Coolhand API key have no rate limit), and as a markdown export for LLM/agent consumption. Filter to a single model with ?q[source_api_eq]=...&q[model_eq]=... — each model's own page shows the exact curl command for that model. The same catalog is also available through the list_inference_apis MCP tool. See https://coolhandlabs.com/docs.