LMArena Text Arena snapshot: 28 Oct 2026
Permanent, dated copy. One view has a board with this date; the other is shown below as missing, never filled in from another date. Live leaderboard →
Download CSV Every row carries the source date, check dates and source link.
Text Arena Overall: default view, as ChatJudge copies it
ChatJudge has no copy of the default view with a page date of 28 Oct 2026. The nearest archived default view board is 27 Oct 2026. It is a separate board from its own source, not merged with this date.
Text Arena Overall: style control off, from Arena’s published dataset
On the Text Arena board Arena published on 28 Oct 2026 (style control off), Gemini 4 Carbon (High) (Google) ranked #1 with an Arena rating of 1540.0 (lower 1530.7, upper 1549.4).
What changed: Since the Arena dataset board published 22 Oct 2026, Gemini 4 Carbon (High) took #1 at 1540.0, replacing Gemini 4 Argon (High) (1525.7 on the previous board); Gemini 4 Carbon (High) entered the top 10 and Muse Spark 1.3 (Max) dropped out.
- Arena and category
- Text Arena, Overall
- Style control
- Off: Arena’s dataset config “text” (on arena.ai this is the “overall-no-style-control” view)
- Source date
- 28 Oct 2026, the leaderboard_publish_date in Arena’s dataset
- Read by ChatJudge
- 15 Nov 2026 (dataset file SHA-256 8f6282c5efc4…)
- Rows
- 31 of 419 models listed. Top 25 overall, plus the best-ranked entry for each major provider outside the top 25. Ratings rounded to 0.1; lower and upper bounds as published in the dataset.
- Same view on arena.ai
- Text Arena, no style control ↗
| Rank | Model | Organisation (as published) | Arena rating | Lower | Upper | Votes |
|---|---|---|---|---|---|---|
| 1 | Gemini 4 Carbon (High) | 1540.0 | 1530.7 | 1549.4 | 4,403 | |
| 2 | Gemini 4 Argon (High) | 1528.6 | 1522.3 | 1534.8 | 11,889 | |
| 3 | Claude Opus 5.5 (High) | anthropic | 1515.8 | 1509.2 | 1522.4 | 11,415 |
| 4 | Claude Fable 5.1 (Max) | anthropic | 1513.8 | 1508.0 | 1519.6 | 17,178 |
| 5 | Claude Opus 5 (Max) | anthropic | 1507.9 | 1503.2 | 1512.6 | 35,531 |
| 6 | Claude Opus 5 (High) | anthropic | 1504.1 | 1500.5 | 1507.6 | 73,167 |
| 7 | Claude Opus 4.6 (High) | anthropic | 1501.7 | 1498.5 | 1505.0 | 84,512 |
| 8 | Muse Wonder (Max) | meta | 1499.2 | 1491.9 | 1506.5 | 7,749 |
| 9 | Gemini 3.8 Flash (High) | 1498.5 | 1494.1 | 1502.9 | 47,803 | |
| 10 | Claude Opus 4.6 | anthropic | 1496.2 | 1493.1 | 1499.4 | 89,324 |
| 11 | Claude Fable 5 (High) | anthropic | 1491.6 | 1487.4 | 1495.7 | 45,341 |
| 12 | MiMo-V2.6-Pro | xiaomi | 1491.4 | 1482.0 | 1500.7 | 4,037 |
| 13 | Muse Spark 1.3 (Max) | meta | 1490.6 | 1485.4 | 1495.8 | 22,134 |
| 14 | Claude Opus 4.7 (High) | anthropic | 1489.7 | 1486.0 | 1493.5 | 65,110 |
| 15 | Gemini 3.7 Flash (High) | 1489.1 | 1484.5 | 1493.8 | 32,289 | |
| 16 | Qwen3.8-Max | alibaba | 1483.3 | 1478.5 | 1488.1 | 31,915 |
| 17 | Claude Opus 4.7 | anthropic | 1483.0 | 1479.2 | 1486.7 | 66,140 |
| 18 | Muse Spark 1.2 (xHigh) | meta | 1482.7 | 1476.3 | 1489.0 | 11,724 |
| 19 | Gemini 3.5 Flash (High) | 1482.2 | 1478.3 | 1486.0 | 48,376 | |
| 20 | Gemini 3.1 Pro Preview | 1480.1 | 1477.2 | 1483.0 | 132,547 | |
| 21 | Gemini 3.6 Flash (High) | 1479.8 | 1475.4 | 1484.2 | 36,028 | |
| 22 | Gemini 3 Pro | 1479.4 | 1475.6 | 1483.2 | 42,057 | |
| 23 | Muse Spark 1.1 | meta | 1478.3 | 1474.2 | 1482.4 | 46,354 |
| 24 | Gemini 3.5 Flash (Medium) | 1478.2 | 1474.0 | 1482.3 | 47,196 | |
| 25 | Kimi K3 (Max) | moonshot | 1475.7 | 1471.3 | 1480.1 | 36,780 |
| 28 | GLM-5.3 (Max) | zai | 1473.8 | 1468.8 | 1478.7 | 26,606 |
| 31 | GPT-5.5 (High) | openai | 1471.6 | 1467.9 | 1475.3 | 69,377 |
| 38 | DeepSeek-V4.1-Flash-20260910 (Max) | deepseek | 1464.5 | 1458.4 | 1470.6 | 13,657 |
| 51 | Grok 4.20 Beta (Reasoning, 0309) | xai | 1451.0 | 1447.3 | 1454.6 | 66,422 |
| 93 | MiniMax-M3 | minimax | 1431.3 | 1427.5 | 1435.0 | 68,611 |
| 95 | Mistral Large 4 | mistral | 1430.1 | 1423.9 | 1436.3 | 12,066 |
Text Arena is run and published by LMArena (lmarena.ai, arena.ai), including the dated leaderboard dataset on Hugging Face. ChatJudge copies the published numbers and does not run the benchmark. Scores from different views or benchmarks aren’t directly comparable.
