TokenRateCalc is a free calculator for estimating what it actually costs to run an application on GPT-5.6, Claude, Gemini, or DeepSeek. No sign-up, no account, no usage limits.
Why This Exists
LLM API pricing is public, but working out a real monthly cost from it isn't simple. Providers price input and output tokens differently, offer different caching and batch discounts, and handle multi-turn conversations differently, where prior responses get resent as context on every new message. Most quick estimates miss at least one of those, and the gap between a naive estimate and a real bill can be substantial.
This calculator does that math properly: real per-model cache and batch rates rather than a flat assumed discount, multi-turn conversation costs that correctly compound, reasoning-token overhead for models that use it, and context-window warnings before you find out the hard way in production.
How Pricing Stays Accurate
Every model's pricing is manually checked against that provider's own official pricing page, not scraped or inferred from secondhand sources. Each entry in the calculator shows its pricing source and the date it was last verified, so you can check the primary source yourself rather than taking it on faith.
This space moves quickly. A number that was accurate last month can be stale today. The "Last Verified" badge in the calculator's header reflects the most recent pass across the full model list, and individual models show their own specific verification date.
What the Calculator Doesn't Do
It estimates model and token costs only. It doesn't account for other charges a provider might bill separately, like tool use, web search or grounding, image or audio generation, embeddings, storage, or fine-tuning. Every number here is an estimate, clearly labeled as one, and the site does not claim more precision than that.
Try the calculator
Free, no sign-up, current pricing across every major providerQuestions or Corrections
If you spot a pricing error or have feedback, get in touch. Corrections get verified against the provider's official source and fixed as quickly as possible.