Arena Leaderboard Dataset
LMArena
Historical snapshots of the Arena leaderboard, showing how model rankings moved over time.
Periodic snapshots of the LMArena leaderboard across categories such as text with style control, with ratings and confidence intervals. One row is a model's rating in a category on a snapshot date. It is updated as the live leaderboard changes.
Access
- Cost
- FreeNo charge.
- License
- Open licenseCC BY 4.0
- Sign-up
- No sign-up
- How you get it
- API and Bulk downloadParquet
What's in it
- One row is
- One row per model, category, and snapshot date
- Covers
- Global users
- Years
- Historical snapshots to the present
- Updated
- Irregular
- Size
- 1 to 10 million rows
Quality
Backed by large-scale human preference votes. Ratings reflect arena voters' preferences, which may not match accuracy.
- Votes reward style and length
- Models can be tuned for the arena
- Provider-submitted private model variants have affected rankings
Connects to
- Chatbot Arena Conversations: Raw votes from an early period of the arena
- Every Eval Ever Datastore: Community database that standardizes leaderboard results across sources
Questions it could answer
- How quickly do new models climb the leaderboard?
- How does style control change the rankings?