All rankings / Search and documents
Documents · style controlled
Human preference ratings for the overall category of this Arena task. Scores from different boards are not interchangeable.
44 entries · Arena score · Higher is better
Source publication: 2026-09-13. Retrieved 2026-11-05.
Source: Arena (lmarena-ai) · CC-BY-4.0.
Ranks use confidence bounds. Several models can share a rank; the rows within a shared rank are displayed by score.
| Significance rank | Model / evaluated variant | Arena score | 95% confidence interval | Votes |
|---|---|---|---|---|
| 1 | claude-fable-5anthropic | 1,498.88 | 1,490.93 – 1,506.82 | 8,497 |
| 1 | claude-opus-4-6-highanthropic | 1,495.26 | 1,488.75 – 1,501.76 | 27,929 |
| 1 | claude-opus-4-7anthropic | 1,494.19 | 1,487.81 – 1,500.57 | 22,192 |
| 1 | claude-opus-4-6anthropic | 1,493.94 | 1,487.99 – 1,499.89 | 41,260 |
| 1 | claude-opus-4-7-highanthropic | 1,493.04 | 1,486.48 – 1,499.59 | 21,957 |
| 1 | claude-fable-5.1-maxanthropic | 1,492.83 | 1,477.55 – 1,508.11 | 1,403 |
| 1 | claude-opus-5-highanthropic | 1,489.75 | 1,482.58 – 1,496.92 | 8,794 |
| 1 | gpt-5.6-sol-xhighopenai | 1,488.67 | 1,479.77 – 1,497.57 | 4,818 |
| 1 | claude-opus-4-8-highanthropic | 1,485.93 | 1,478.98 – 1,492.89 | 11,551 |
| 2 | gpt-5.5openai | 1,482.81 | 1,476.45 – 1,489.17 | 21,279 |
| 5 | gpt-5.5-highopenai | 1,480.24 | 1,473.74 – 1,486.75 | 20,944 |
| 3 | muse-spark-1.1meta | 1,479.46 | 1,470.51 – 1,488.41 | 4,885 |
| 6 | claude-sonnet-4-6anthropic | 1,477.73 | 1,471.78 – 1,483.68 | 57,813 |
| 6 | gpt-5.6-terra-xhighopenai | 1,476.65 | 1,468.33 – 1,484.98 | 5,787 |
| 6 | claude-sonnet-5-highanthropic | 1,475.7 | 1,468.03 – 1,483.36 | 7,411 |
| 7 | claude-opus-4-8anthropic | 1,475.38 | 1,468.33 – 1,482.43 | 12,128 |
| 5 | gpt-6-astra-maxopenai | 1,472.56 | 1,458.27 – 1,486.85 | 1,621 |
| 11 | gpt-5.4openai | 1,470.09 | 1,463.82 – 1,476.36 | 33,331 |
| 6 | muse-spark-1.3-maxmeta | 1,467.89 | 1,449.55 – 1,486.22 | 1,006 |
| 6 | muse-sparkmeta | 1,467.29 | 1,448.93 – 1,485.66 | 1,084 |
| 11 | claude-opus-4-5-20251101anthropic | 1,464.61 | 1,454.23 – 1,474.99 | 7,981 |
| 12 | grok-4.5xai | 1,464.35 | 1,455.55 – 1,473.14 | 4,965 |
| 14 | gpt-5.6-luna-xhighopenai | 1,461.98 | 1,453.6 – 1,470.37 | 5,766 |
| 12 | grok-4.6-highxai | 1,461.07 | 1,449.01 – 1,473.14 | 2,258 |
| 14 | gemini-3.5-flash-mediumgoogle | 1,460.56 | 1,452.63 – 1,468.49 | 6,500 |
| 14 | gpt-5.5-instantopenai | 1,460.32 | 1,451.51 – 1,469.13 | 8,497 |
| 17 | gemini-3.1-pro-previewgoogle | 1,458.86 | 1,453.56 – 1,464.17 | 49,457 |
| 17 | gemini-3.6-flash-highgoogle | 1,453.42 | 1,442.51 – 1,464.33 | 2,975 |
| 18 | gemini-3.5-flash-highgoogle | 1,453.4 | 1,444.04 – 1,462.75 | 4,061 |
| 18 | qwen3.7-plusalibaba | 1,452.54 | 1,443.46 – 1,461.63 | 4,591 |
| 18 | gemini-3-progoogle | 1,451.09 | 1,442.13 – 1,460.05 | 10,769 |
| 19 | kimi-k2.6moonshot | 1,449.8 | 1,442.07 – 1,457.53 | 11,291 |
| 20 | claude-sonnet-4-5-20250929anthropic | 1,449.08 | 1,442.85 – 1,455.3 | 31,455 |
| 24 | gemma-4-31bgoogle | 1,444.51 | 1,436.65 – 1,452.36 | 12,326 |
| 28 | grok-4.20-beta-0309-reasoningxai | 1,439.43 | 1,432.33 – 1,446.53 | 20,948 |
| 28 | gemini-3-flashgoogle | 1,439.15 | 1,429.78 – 1,448.52 | 7,158 |
| 31 | kimi-k2.5-thinkingmoonshot | 1,435.96 | 1,429.15 – 1,442.78 | 21,569 |
| 34 | minimax-m3minimax | 1,432.96 | 1,424.64 – 1,441.29 | 6,314 |
| 34 | gemini-2.5-progoogle | 1,430.58 | 1,424.21 – 1,436.95 | 25,110 |
| 36 | gpt-5.2openai | 1,425.37 | 1,419.38 – 1,431.35 | 28,212 |
| 35 | gpt-5.2-highopenai | 1,423.16 | 1,413.69 – 1,432.64 | 7,108 |
| 38 | claude-haiku-4-5-20251001anthropic | 1,418.73 | 1,412.76 – 1,424.7 | 34,677 |
| 38 | gpt-5.1openai | 1,415.84 | 1,406.54 – 1,425.14 | 8,244 |
| 43 | glm-5v-turbozai | 1,400.07 | 1,391.66 – 1,408.47 | 6,995 |
What this ranking can tell you
- The population is the models in this source board, including variants and models without an available API price.
- Ranks use the published confidence bounds: one plus the number of models whose lower bound exceeds this model’s upper bound. Shared ranks are not proof of equal performance.
Check the source
The original evidence is retained byte for byte. The downloadable board includes the source URLs, revision and SHA-256 hashes.
Download this board and provenance →