Home / AI API cost calculator

AI API cost calculator

Enter how many tokens you send and receive, and how often. See what it costs each month on every model we track.

Prices checked

Your workload

Tokens per month

Input
100M
Output
30M
Total
130M

Cheapest for this workload

 

 

 a month

 

Open weightsCheapest you can also run yourself
  

 

Every model, cheapest first

The same workload priced on all 25 models we track, at each lab’s list price per million tokens.

 

Monthly cost of this workload on each model, cheapest first
Rank Model Price per 1M tokensinput / output Cost per request Monthly cost vs cheapest

How the cost is calculated

One sum per model, using list prices checked on November 18, 2026.

Monthly cost = requests × (input tokens × input price + output tokens × output price) ÷ 1,000,000

  • Long prompts cost more on four models. When input per request passes the threshold (for Grok, reaches it), the higher rate applies to the whole request: Claude Haiku 5.5 above 100K tokens ($0.50 / $2.50), Grok 4.7 from 200K ($4 / $12), Doubao Seed 2.0 Pro above 128K ($1 / $6) and MiniMax M3 above 512K ($0.60 / $2.40).
  • Prompts that do not fit are not priced. If your input is larger than a model’s context window, the row says “Prompt too long”. Xiaomi does not state a context window for MiMo-v2.6 Pro, so it is always priced; check that your prompt fits.
  • DeepSeek off-peak halves both DeepSeek prices. Turn it on if your jobs can run outside peak hours.
  • Gemini 4 Argon and Carbon cost $2 / $10 until November 30, 2026, then $4 / $20. The switch applies the later price.
  • Not included: prompt caching, batch discounts, free tiers, and taxes. Your bill can be lower or higher than the figure shown.

Questions people ask

How do I estimate tokens per request?

In English, 1,000 tokens is roughly 750 words, though the count varies by model and language. Input is everything you send each time: instructions, chat history, and any documents or code. Output is the reply. If you already use an API, the usage numbers in its responses give you exact figures.

Why does a model show “Prompt too long”?

The input tokens you entered are more than that model’s context window, so it cannot take the request in one go. Mistral Medium 3.5, Doubao Seed 2.0 Pro and Hy3 stop at 256K tokens and Grok 4.7 at 500K. Most other models here take about 1M.

Will my real bill match this number?

It will be close if your requests are all about the same size. The calculator uses list prices and leaves out prompt caching, batch discounts and free tiers. Mistral’s batch processing costs 50% less and OpenAI’s background requests 30% to 40% less, so heavy batch jobs can come in well under the figure shown.