Tool · 62 models · prices checked Sep 16, 2026
LLM API cost calculator
Describe a typical request and your monthly volume, and compare what every model would cost. Not sure how many tokens your prompt is? Paste it into the token counter first.
The calculator does not check context windows: a model only appears affordable for long prompts if it can accept them. Check limits on the model pages.
| # | Model | Vendor | Per request | Per month | Applied |
|---|---|---|---|---|---|
| Enable JavaScript to run the calculator, or see static prices on the LLM API pricing page. | |||||
How to estimate your inputs
Input tokens are everything you send on each call: system prompt, conversation history, retrieved documents and the user's message. In a chat app history grows with every turn, so use an average across a conversation, not the first message. Output tokens are what the model writes back, plus hidden reasoning tokens on reasoning models, which vendors bill as output.
Cached input is the share of each prompt that repeats exactly from a previous request, typically a long fixed system prompt or document placed at the start. If your system prompt is 3,000 tokens of a 4,000-token request, set the slider to 75%. The calculator ignores cache-write charges, which OpenAI and Anthropic list for some models, so real savings there are slightly lower.
Batch processing suits work that can wait, such as classification, evaluation runs or nightly summaries. Vendors return batch results within hours instead of seconds and charge less for them.
Reading the result
The table is sorted from cheapest to most expensive for your settings. The "Applied" column shows which pricing rules changed a model's rate: long-context pricing when your input passes a threshold, batch or cached prices, peak hours, and promotional prices that will end. The cheapest model is not automatically the right one: test a few candidates on your own prompts, then check full rate cards on the LLM API pricing pages.
Questions
How is the cost calculated?
For each model: uncached input tokens × input price + cached input tokens × cached input price + output tokens × output price, divided by 1,000,000 because prices are per million tokens. That gives the cost per request, which is multiplied by requests per month.
Which prices does the calculator use?
The list prices on each vendor's own pricing page, checked on Sep 16, 2026: standard processing in US dollars. Long-context rates apply automatically when input tokens per request exceed a model's threshold. DeepSeek uses off-peak prices unless you tick peak hours.
What does the batch option do?
It applies batch prices where the vendor lists them: every Anthropic model, several Gemini models, and OpenAI, which states a 50% Batch API discount. Models without a published batch price keep their standard price and are marked in the table. Cached input stays at its normal cached rate, so batch estimates with caching are slightly conservative.
Why might my real bill be different?
Token counts differ between vendors for the same text, reasoning models bill thinking tokens as output, OpenAI and Anthropic charge extra to write to the prompt cache on some models, and features such as web search, images and data residency have separate prices. Use the result to compare models, then confirm on the vendor page.