Figures on this page were verified 31 August 2026 against the providers' own documentation. Pricing, context windows and rate limits change without notice, so confirm any number against the provider before you rely on it. Tell us if something here is out of date.
Every current model from Anthropic, OpenAI and Google in one sortable table: context window, maximum output, and price per million tokens. Sort by any column to find the cheapest model that still fits your workload. Everything runs in your browser and every figure links back to the provider documentation it was read from.
Blended $/M weights input and output 3:1, which is closer to a real chat workload than either price alone. An em-dash means the provider does not publish that figure on the page we verified, and we would rather show a gap than a guess.
How to read this table
The headline input price is the least useful number on the page, because almost no workload is input-only. Output tokens cost between three and six times more than input on every provider here, so a model that looks cheap on input can be the expensive option once it starts writing.
That is what the blended column is for. It weights input and output 3:1, roughly the shape of a chat or retrieval workload. If your usage is code generation or long-form writing the real ratio is closer to 1:1, which pushes expensive-output models further down the list. If you are summarising large documents it is closer to 10:1 and input price dominates. The API cost calculator lets you set your own volumes rather than accepting our assumption.
Context window is not the whole story
A million-token context window does not mean you can send a million tokens of input. The reply is reserved from the same budget, so a request with max_tokens set to 128,000 has that much less room before it starts. This is the single most common cause of context_length_exceeded, and the context budget planner exists to make the arithmetic visible before you hit it.
Maximum output matters independently. A model with a huge context but a modest output ceiling will truncate long answers with stop_reason: max_tokens, which returns HTTP 200 and is therefore easy to mistake for success. That failure is covered in why the response got cut off.
Discounts this table does not show
The prices here are the standard synchronous rates. Several mechanisms cut them substantially, and none are reflected above because they depend on how you call the API rather than which model you pick.
- Batch processing is 50% off on Anthropic for work that tolerates delayed completion.
- Prompt caching charges cache reads at 10% of the base input price on Anthropic, which is transformative for a long fixed system prompt reused across many turns.
- Promotional pricing applies to some models on a stated deadline. Gemini 3.7 Flash is on a promotional rate through 31 December 2026, and the table flags it.
- Long-context surcharges apply on some providers above a threshold, so the headline rate is a floor rather than a flat price.
Where these numbers come from
Every figure was read from the provider’s own documentation on the date shown at the top of this page: Anthropic’s models overview, OpenAI’s models reference, and Google’s Gemini API pricing page. Nothing here is copied from another comparison site or written from memory.
Google publishes pricing and context limits on separate pages and does not state a context window for every model on the pricing page we verified, which is why several Google rows show a dash rather than a number. Providers change prices without notice, so treat this as a starting point and confirm before committing to a budget. If something here has gone stale, tell us and we will fix the page and update its date.
Frequently asked questions
Why do some rows show a dash instead of a number?
Because the provider does not publish that figure on the page we verified. Google lists pricing and context limits separately and does not state a window for every model on its pricing page. We would rather show a visible gap than fill it with a guess that looks authoritative.
What does the blended price column mean?
It weights input and output tokens 3:1, which is closer to a real chat or retrieval workload than either price on its own. Output costs three to six times more than input everywhere, so a model that looks cheap on input can be the expensive choice once it starts writing.
Does a one million token context window mean I can send a million tokens?
No. The reply is reserved from the same budget, so setting max_tokens to 128,000 leaves that much less room for input before the request is even sent. This is the most common cause of context_length_exceeded.
Why are batch and caching discounts not in the table?
Because they depend on how you call the API rather than which model you choose. Batch processing is 50% off on Anthropic, and cached input reads cost 10% of the base rate, which changes the ranking completely for a long reused system prompt.