How to use it
- Enter a typical request: input tokens, output tokens and how many requests you expect.
- Search or filter by provider to narrow the list.
- Select a column heading to sort. "Your workload" shows the total cost for the numbers you entered.
How to read this table
Prices are in US dollars per one million tokens, for standard pay-as-you-go API access. Input is what you send: instructions, documents and chat history. Output is what the model writes back. Cached input is the discounted rate some providers charge when the start of your prompt is identical to a recent request. Context is the largest number of tokens the model can take in one request.
Why output prices matter more than they look
Output tokens typically cost four to eight times as much as input tokens. For a chatbot that reads short questions and writes long answers, output dominates the bill. For a classifier that reads long documents and replies with one word, input dominates. That is why this page lets you enter your own token mix: the ranking changes depending on what you do.
Worked example
A support assistant handles 30,000 questions a month. Each request sends about 2,000 tokens (instructions plus the question) and gets back about 400 tokens. That is 60 million input and 12 million output tokens a month. On a model priced at $2 input and $10 output, the monthly cost is $120 + $120 = $240. On a model at $0.30 input and $1.20 output, the same work costs $18 + $14.40 = $32.40. Type those numbers into the workload boxes above to see the full ranking.
How these prices are kept current
Every week an automated job reads a public, machine-readable price list maintained by the open-source LiteLLM project, picks out the main models from each provider, and updates this page. If a price changes by more than three times in a single week, the change is held back for a person to check instead of being published. You can see every change on the data updates page.
Prices change, discounts apply for batch jobs and committed spend, and some providers charge more for very long prompts. Treat this table as a starting point and confirm on the provider's pricing page, linked under the table, before you commit a budget.
Other costs to plan for
- Batch processing. Most major providers discount jobs that can wait up to a day, often by around half. Model this with the batch cost calculator.
- Long-context surcharges. Some models charge a higher rate once a prompt passes a size threshold.
- Tools and search. Web search, code execution and image inputs can be billed separately from tokens.
- Reasoning tokens. Thinking models bill their hidden reasoning as output tokens, so real output can be several times the visible answer.
Questions
How often are these prices updated?
An automated check runs every week. The date of the latest change is shown above the table, and every change is listed on the data updates page.
Where does the data come from?
From the public model price list maintained by the open-source LiteLLM project, which many developer tools rely on. We filter it to the main models from each provider and hold back any suspiciously large change for manual review.
Are these prices in US dollars?
Yes. All prices are USD per one million tokens, before tax and before any volume, batch or enterprise discounts.
Why is a model I use missing?
We show the main current text models from six providers and leave out dated snapshots, preview builds, fine-tunes and audio or image-only models to keep the list readable.