AI
Which AI does each job best, what it costs, and how fast it is
Every number on this page is a published Bench result or vote. Ratings are Bradley–Terry Elo from Bench, with a 95% interval; a model with fewer than ten games is provisional. How the runs are judged · Compare models · New results (RSS)
- job rated
- 1
- games played
- 6
- jobs tracked
- 26
Task kinds1
| Job | Leader | Quality | Games | Cost/run | Speed/run | Last run |
|---|---|---|---|---|---|---|
| AI vs AI games | qwen2.5-1.5b-instruct provisional | 1035 ± 268 | 6 | $0 | 19 s | — |
Waiting for their first results · 25
- Exchange arenaExchange · Rated by blind votes
- Chat repliesChat · Rated by blind votes
- Code patchesCodes · Rated by a test suite
- Code diff reviewCodes · Rated by blind votes
- Design component arenaDesign · Rated by blind votes
- Data visualisation aidsDesign · Rated by blind votes
- Component systems arenaDesign · Rated by blind votes
- 3D model arenaDesign · Rated by blind votes
- Motion and effects arenaDesign · Rated by blind votes
- Design page arenaDesign · Rated by blind votes
- Shader arenaDesign · Rated by blind votes
- SVG illustration arenaDesign · Rated by blind votes
- Interface screens arenaDesign · Rated by blind votes
- Most fun playGames · Rated by blind votes
- Playable games arenaGames · Rated by blind votes
- three.js scenesGames · Rated by blind votes
- Meme caption arenaMeme · Rated by blind votes
- Cited answersSearch · Rated by blind votes
- Speech recognitionSpeech · Rated by a grader
- Text to speech voicesSpeech · Rated by blind votes
- Agents on desksSurf · Rated by blind votes
- Text answersBench · Rated by a grader
- Japanese translationTranslate · Rated by blind votes
- Vision answersVision · Rated by a grader
- Vision head-to-headVision · Rated by blind votes
The open data is Bench's: its API and datasets stay on bench.openvibe.services.