Compare AI models
Put up to four models side by side: tier and Elo on every leaderboard, prices, context window and release date.
Updated · Sources: models.dev
| Metric | GPT-4o (2024-08-06) OpenAI |
|---|---|
| Text | B 1,282 1,278–1,287 |
| Coding | B 1,284 1,276–1,291 |
| Input / output, per 1M | $2.50 / $10 |
| Context window | 128K |
| Max output | 16K |
| Released | August 6, 2024 |
| Knowledge cutoff | 2023-09 |
| License | — |
| Input | text, image |
| App | — |
What the comparison shows
Put up to 4 models next to each other and see, column by column, how they stand: tier and rating on every leaderboard where the model is ranked, prices per unit, context window, maximum output, supported modalities, whether the weights are open, and the release date. The ratings come from LMArena, the prices and technical data from models.dev and LiteLLM.
How to pick models
- Start typing a name in one of the fields and choose from the suggestions.
- Or tick the compare box in any leaderboard table; a bar at the bottom collects your choice and opens this page.
The address of a comparison contains the chosen models, so you can share the link. Only the empty comparison page is indexed by search engines.
Reading the table correctly
Compare ratings within one board. An Elo of the text board and the score of the agents board are different scales, and even two Elo boards are separate arenas with separate voters. A model that leads in image generation says nothing about its rank in coding.
An empty cell means that the source has no value for that model — no price at any provider, no known knowledge cutoff — not that the value is zero. Prices are the cheapest current offers on the day of the data; the details of each model on the leaderboard pages list the other providers too.
When a comparison helps most
It is useful at the end of a choice rather than at the start. Use the leaderboards and the quality-versus-price chart to narrow the field, then compare the finalists on the things that matter for your case: context length for long documents, modalities if you need images or audio, open weights if you want to run the model yourself.