Providers price input and output tokens separately — and output usually costs 3–5× more than input. So a short prompt with a long answer can cost more than a long prompt with a short answer. Enter your two numbers and see the real per-call cost across models.
—cheapest model
—total for calls
—tokens / call
Cost per call by model
Your token counts, every model, ranked cheapest first. Prices are per 1M tokens.
| Model | In $/1M | Out $/1M | Cost / call |
|---|
⚠️ Estimate using reference prices (July 2026) and list rates. Actual prices change and vary by tier, region, context length and caching/batch discounts. For repeated context, see prompt caching; for bulk async jobs, see batch pricing. ·
Report outdated price →
How token cost works
The formula is simple: cost = (input tokens ÷ 1,000,000 × input price) + (output tokens ÷ 1,000,000 × output price). The two prices are different on purpose — generating text (output) is more expensive than reading it (input), typically by 3–5×. That's why the biggest lever on your bill is usually capping output length, not trimming the prompt. As a rough guide, 1,000 tokens ≈ 750 English words, but code and other languages tokenize differently, so measure with the provider's tokenizer for exact figures.
Next steps
Estimating a whole product, not one call? Use the AI API cost calculator or the AI app cost estimator. Building an agent or chatbot? Try the AI agent cost calculator or chatbot cost calculator. Reusing the same context every call? Prompt caching and batch jobs each cut cost further.
OpenAIBinanceTwilioGoogle GeminiCoinGecko
How this calculator works
The LLM Token Cost Calculator estimates what a single language-model API call costs based on the number of input tokens you send and the output tokens the model returns. Because providers price input and output tokens at different per-million rates, both counts drive the total, and output tokens are usually the more expensive of the two. You can also enter a call count to project the cost across many identical requests. The tool applies each model's published rates and then ranks the results cheapest first across providers such as GPT, Claude, Gemini, and DeepSeek, so you can compare the same workload side by side.
The key trade-off to watch is that the lowest per-token price is not always the lowest real cost. A model that produces longer, more verbose responses can burn more output tokens and end up pricier than a nominally more expensive one. Estimate with token counts that reflect your actual prompts and typical response length, since a short test rarely predicts production volume. Re-check the ranking whenever your prompt size, expected output, or call frequency changes, because the cheapest option can shift as any of those inputs move.
Frequently asked questions
How is LLM token cost calculated?
Cost = (input tokens / 1,000,000 × input price) + (output tokens / 1,000,000 × output price). Providers price input and output separately, and output is usually 3-5x more expensive than input, so long answers cost more than long prompts.
How many tokens is a word?
For English, roughly 1 token is about 0.75 words, so 1,000 tokens is around 750 words. Code, numbers and other languages tokenize differently, so always measure with the provider's tokenizer for exact counts.