Industry · Sep 10, 2026
What counts as a release on this site
A versioned model, a documented API change, or a serving stack people can actually run. Not a keynote vibe.
Live constellation · Sep 2026 cut
The live map of who is winning each orbit.
Category leaderboards for chat, coding, image, video, and open-weight models — scored from public arenas, then a desk that writes when something actually ships.
Text arena · Elo
Chat
Updated 2026-09-02
| Rank | Model | Elo | Notes |
|---|---|---|---|
| 01 | Claude Fable 5Anthropic | 1507 | 27k votes · $10 / $50 |
| 02 | Claude Opus 4.6 highAnthropic | 1505 | 72k votes · $5 / $25 |
| 03 | Claude Fable 5.1 maxAnthropic | 1504 | 2.9k votes · thin sample |
| 04 | Claude Opus 4.7 highAnthropic | 1502 | 60k votes · $5 / $25 |
| 05 | Muse Spark 1.2 xHighMeta | 1499 | 3.2k votes · $1.25 / $4.25 |
| 06 | Claude Opus 4.6Anthropic | 1498 | 76k votes · default effort |
| 07 | Gemini 3.8 Flash highGoogle | 1494 | 5.1k votes · preliminary |
| 08 | Claude Opus 5 highAnthropic | 1493 | 35k votes |
Blind pairwise human preference on arena.ai (LMArena text). Style can beat rigor. Rankings move as votes land.
arena.ai text leaderboardPrefer a permalink? Open the chat board.
Recent signals
What actually shipped
Industry · Sep 10, 2026
How this desk covers model releases
We read primary sources, write only when something actually shipped, and leave rumors on the floor.