LLM API Pricing (2026)

LLM APIs bill per token processed — roughly 4 characters or 0.75 English words each. Input tokens are what you send (prompts, documents), output tokens are what the model writes back. Output is typically 3–6× more expensive than input.

Generating tokens is far more compute-intensive than reading them: every output token requires a full forward pass through the model, while input tokens are processed in parallel. Vendors price accordingly.

By raw input price, small open-weight models hosted by inference providers usually win, followed by vendor "mini/nano/flash" tiers. But cheapest per token is not cheapest per task — a stronger model may need fewer retries and shorter prompts.

Every row shows a verification date and links to the official pricing page it was checked against. We refresh the dataset periodically; always confirm on the vendor page before committing to volume.