How we make the score
- Agentic
- 20%
- Coding
- 20%
- Reasoning
- 15%
- Multimodal
- 10%
- Knowledge
- 15%
- Multilingual
- 5%
- Instruction
- 5%
- Math
- 10%
Methodology
Publishers we read
- Vals AI
- 3,176 results.
- EuroEval
- 1,400 results.
- Epoch AI
- 1,168 results.
- ARC Prize Foundation
- 451 results.
- LiveBench
- 406 results.
- LMArena
- 132 results.
- SWE-bench team
- 109 results.
- Sierra Research
- 90 results.
- SWE-bench team
- 82 results.
- Terminal-Bench
- 15 results.
- OpenRouter
- 444 model records.