Model Rankings
These rankings help you compare the relative performance of different models; they should not be the only input to a model choice. The data comes from Arena AI and is refreshed daily by GitHub Actions.
When choosing a model, also weigh the task type, context length, multimodal and tool-calling abilities, speed, price, regional availability and data policy. Rankings and prices change over time: trust the update time shown on each page and the model provider's own documentation.
Leaderboards
- Agent Leaderboard: This leaderboard measures how models perform on multi-turn tool-calling and agent tasks, across net improvement, confirmed success, praise-to-complaint ratio, steerability, Bash recovery, tool hallucination and session volume.
- Text Leaderboard: This is Arena AI's core text battle leaderboard, which estimates relative model strength from blind human-voted matchups.
- Search Leaderboard: This leaderboard measures how models perform on web search and information-retrieval tasks.
- Vision Leaderboard: This leaderboard measures how multimodal models perform on image-understanding tasks.
- Code / Web Dev Leaderboard: This leaderboard measures how models perform in practice on web front-end development (HTML/CSS/JS) tasks.
- Text-to-Image Leaderboard: This leaderboard measures how well text-to-image models turn a text prompt into an image.
💡 Get help and send feedback
If you run into questions, bugs or feature suggestions while configuring or using BLACK HOLE, please use the official channels listed in Feedback and suggestions.