Benchmarks
External leaderboards from Arena, for the models on this site. These are not scores for our six prompts, which still get no score and no ranking.
Arena shows two anonymous answers and asks which is better, then turns the votes into a rating and a rank. A rating is a relative strength score, not a percentage or a grade: it only means something beside the other ratings in the same table.
We match a model here to Arena's name for the same model. Arena adds the reasoning effort to some names (“Claude Opus 5.5 (Max)”), so the exact matched name is in each table. When Arena has no entry for a model, it has no row here: no number is guessed or copied from a sibling model. Ranks are Arena's place among every model it ranks, not a place among the models on this site, so gaps are expected.
Code Arena
Web pages and interfaces, built from a prompt and voted blind. Arena's leader: Claude Fable 5.2 (xHigh) (1,824).
| Rank | Model | Rating | Votes | Arena name |
|---|---|---|---|---|
| #2of 160 | Claude Opus 5.5Anthropic | 1,818 | 4,792 | Claude Opus 5.5 (Max) |
| #5of 160 | Gemini 4 CarbonGoogle | 1,769 | 2,092 | Gemini 4 Carbon (High) |
| #6of 160 | Gemini 4 NeonGoogle | 1,764 | 2,310 | Gemini 4 Neon (High) |
| #8of 160 | GPT-6.2 SolOpenAI | 1,757 | 1,198 | GPT-6.2 Sol (Max) |
| #9of 160 | Qwen4 MaxAlibaba | 1,750 | 1,811 | Qwen4-Max (xHigh) |
| #10of 160 | GPT-6.1 SolOpenAI | 1,744 | 6,285 | GPT-6.1 Sol (Max) |
| #11of 160 | Claude Fable 5.1Anthropic | 1,741 | 10,458 | Claude Fable 5.1 (Max) |
| #13of 160 | Claude Sonnet 5.5Anthropic | 1,716 | 8,666 | Claude Sonnet 5.5 (High) |
| #14of 160 | Gemini 4 HydrogenGoogle | 1,703 | 2,040 | Gemini 4 Hydrogen (High) |
| #16of 160 | Kimi K3.1 ProMoonshot AI | 1,684 | 5,799 | Kimi K3.1 Pro (Max) |
| #17of 160 | GPT-6.1 TerraOpenAI | 1,682 | 4,972 | GPT-6.1 Terra (Max) |
| #18of 160 | Gemini 4 ArgonGoogle | 1,681 | 5,784 | Gemini 4 Argon (High) |
| #19of 160 | Qwen3.8 Max (0902)Alibaba | 1,675 | 24,725 | Qwen3.8-Max (0902) |
| #21of 160 | Qwen3.8 Max PrimeAlibaba | 1,671 | 3,360 | Qwen3.8-Max |
| #22of 160 | DeepSeek V4.1 Flash 1017DeepSeek | 1,659 | 17,328 | DeepSeek-V4.1-Flash-20261017 (Max) |
| #23of 160 | Kimi K3Moonshot AI | 1,658 | 23,171 | Kimi K3 (Max) |
| #25of 160 | Muse Spark 1.3Meta | 1,654 | 13,137 | Muse Spark 1.3 (Max) |
| #26of 160 | Muse WonderMeta | 1,648 | 6,299 | Muse Wonder (Max) |
| #27of 160 | Inkling 2 SmallThinking Machines | 1,644 | 1,185 | Inkling 2 Small |
| #28of 160 | Grok 4.7xAI | 1,639 | 9,078 | Grok 4.7 (xHigh) |
| #29of 160 | MiMo v2.6 FlashXiaomi | 1,636 | 9,802 | MiMo-V2.6-Flash |
| #30of 160 | Hy4 PreviewTencent | 1,631 | 4,578 | Hy4 preview |
| #33of 160 | MiMo v2.6 ProXiaomi | 1,623 | 9,978 | MiMo-V2.6-Pro |
| #34of 160 | GLM 5.3 PrimeZ.ai | 1,622 | 13,566 | GLM-5.3 (Max) |
| #34of 160 | GLM 5.3Z.ai | 1,622 | 13,566 | GLM-5.3 (Max) |
| #36of 160 | DeepSeek V4.1 FlashDeepSeek | 1,621 | 3,378 | DeepSeek-V4.1-Flash-20260910 (Max) |
| #39of 160 | GLM 5.3 FlashZ.ai | 1,610 | 24,528 | GLM-5.3-Flash |
| #41of 160 | Muse EmberMeta | 1,608 | 1,327 | Muse Ember (Max) |
| #43of 160 | Claude Haiku 5.5Anthropic | 1,596 | 8,080 | Claude Haiku 5.5 (High) |
| #46of 160 | Gemini 3.8 FlashGoogle | 1,585 | 14,882 | Gemini 3.8 Flash (High) |
| #47of 160 | MiniMax M3.1 FlashMiniMax | 1,582 | 5,732 | MiniMax-M3.1-Flash |
| #50of 160 | GPT-6 LunaOpenAI | 1,578 | 31,984 | GPT-6 Luna (Max) |
| #51of 160 | Step 5 PreviewStepFun | 1,571 | 8,338 | Step 5 Preview (High) |
| #52of 160 | Mistral Large 4Mistral | 1,568 | 2,566 | Mistral Large 4 |
| #64of 160 | Muse Spark OpenMeta | 1,527 | 1,991 | Muse Spark Open (Max) |
| #79of 160 | MiniMax M3MiniMax | 1,484 | 20,957 | MiniMax-M3 |
| #94of 160 | Beam 1.0Reflection AI | 1,437 | 2,502 | Beam 1.0 501B A23B |
| #103of 160 | InklingThinking Machines | 1,414 | 17,097 | Inkling |
| #107of 160 | Nemotron 3.5 Super VLNVIDIA | 1,402 | 6,007 | NVIDIA-Nemotron-3.5-Super-VL-120B-A12B (NVFP4) |
Arena's table of 18 Nov 2026.
Text Arena
General chat, writing and knowledge, with style control on. Arena's leader: Gemini 4 Carbon (High) (1,526).
| Rank | Model | Rating | Votes | Arena name |
|---|---|---|---|---|
| #1of 422 | Gemini 4 CarbonGoogle | 1,526 | 4,266 | Gemini 4 Carbon (High) |
| #2of 422 | Gemini 4 ArgonGoogle | 1,521 | 25,751 | Gemini 4 Argon (High) |
| #3of 422 | Gemini 4 NeonGoogle | 1,518 | 3,885 | Gemini 4 Neon (High) |
| #4of 422 | Muse WonderMeta | 1,511 | 15,331 | Muse Wonder (Max) |
| #5of 422 | Claude Opus 5.5Anthropic | 1,507 | 17,727 | Claude Opus 5.5 (High) |
| #8of 422 | Claude Fable 5.1Anthropic | 1,502 | 21,941 | Claude Fable 5.1 (Max) |
| #10of 422 | Muse Spark 1.3Meta | 1,498 | 29,406 | Muse Spark 1.3 (Max) |
| #11of 422 | Gemini 3.8 FlashGoogle | 1,497 | 70,822 | Gemini 3.8 Flash (High) |
| #16of 422 | Gemini 4 HydrogenGoogle | 1,490 | 4,737 | Gemini 4 Hydrogen (High) |
| #21of 422 | Kimi K3Moonshot AI | 1,487 | 43,641 | Kimi K3 (Max) |
| #23of 422 | Kimi K3.1 ProMoonshot AI | 1,487 | 6,011 | Kimi K3.1 Pro (Max) |
| #26of 422 | Qwen3.8 Max PrimeAlibaba | 1,484 | 39,457 | Qwen3.8-Max |
| #27of 422 | GPT-6.1 SolOpenAI | 1,483 | 15,324 | GPT-6.1 Sol (Max) |
| #29of 422 | GLM 5.3 PrimeZ.ai | 1,483 | 34,014 | GLM-5.3 (Max) |
| #29of 422 | GLM 5.3Z.ai | 1,483 | 34,014 | GLM-5.3 (Max) |
| #32of 422 | MiMo v2.6 ProXiaomi | 1,480 | 4,056 | MiMo-V2.6-Pro |
| #38of 422 | GLM 5.3 FlashZ.ai | 1,475 | 54,903 | GLM-5.3-Flash |
| #41of 422 | DeepSeek V4.1 FlashDeepSeek | 1,475 | 13,505 | DeepSeek-V4.1-Flash-20260910 (Max) |
| #47of 422 | Claude Sonnet 5.5Anthropic | 1,474 | 16,356 | Claude Sonnet 5.5 (xHigh) |
| #85of 422 | Step 5 PreviewStepFun | 1,449 | 12,298 | Step 5 Preview |
| #88of 422 | MiMo v2.6 FlashXiaomi | 1,448 | 27,512 | MiMo-V2.6-Flash |
| #94of 422 | Grok 4.7xAI | 1,443 | 24,710 | Grok 4.7 (xHigh) |
| #96of 422 | InklingThinking Machines | 1,442 | 53,382 | Inkling |
| #98of 422 | GPT-6 LunaOpenAI | 1,441 | 115,874 | GPT-6 Luna (Max) |
| #109of 422 | MiniMax M3MiniMax | 1,435 | 75,997 | MiniMax-M3 |
| #111of 422 | Beam 1.0Reflection AI | 1,434 | 3,990 | Beam 1.0 501B A23B |
| #122of 422 | Nemotron 3 UltraNVIDIA | 1,426 | 11,376 | NVIDIA-Nemotron-3-Ultra-550B-A55B (NVFP4) |
| #133of 422 | Mistral Large 4Mistral | 1,422 | 9,078 | Mistral Large 4 |
| #134of 422 | Claude Haiku 5.5Anthropic | 1,421 | 16,620 | Claude Haiku 5.5 (High) |
| #144of 422 | Nemotron 3.5 Super VLNVIDIA | 1,415 | 4,040 | NVIDIA-Nemotron-3.5-Super-VL-120B-A12B (NVFP4) |
| #152of 422 | Inkling 2 SmallThinking Machines | 1,412 | 2,678 | Inkling 2 Small |
| #222of 422 | Nemotron 3.5 LightningNVIDIA | 1,352 | 25,044 | NVIDIA-Nemotron-3.5-Lightning-30B-A3B (NVFP4) |
Arena's table of 18 Nov 2026.
Image Arena
Pictures made from a text prompt. Arena's leader: GPT-Image-2.5 Sunburst (1,428).
| Rank | Model | Rating | Votes | Arena name |
|---|---|---|---|---|
| #5of 86 | GPT-5.4 Image 2OpenAI | 1,382 | 110,696 | GPT-Image-2 (Medium) |
| #8of 86 | Nano Banana 2.1Google | 1,328 | 4,957 | Gemini Nano Banana 2.1 |
| #14of 86 | Nano Banana 2Google | 1,256 | 84,052 | Gemini 3.1 Flash Image (nano-banana-2, Web Search) |
| #19of 86 | Nano Banana ProGoogle | 1,246 | 184,617 | Gemini 3 Pro Image (nano-banana-pro, 2K) |
Arena's table of 17 Nov 2026.
Video Arena
Clips made from a text prompt. Arena's leader: Gemini Omni 1.2 Flash (1,540).
| Rank | Model | Rating | Votes | Arena name |
|---|---|---|---|---|
| #1of 50 | Gemini Omni 1.2 FlashGoogle | 1,540 | 1,714 | Gemini Omni 1.2 Flash |
| #2of 50 | Gemini Omni 1.1 FlashGoogle | 1,513 | 1,792 | Gemini Omni 1.1 Flash |
Arena's table of 9 Nov 2026.
Image-to-video Arena
Clips made from a picture and a prompt. Arena's leader: Gemini Omni 1.2 Flash (1,503).
| Rank | Model | Rating | Votes | Arena name |
|---|---|---|---|---|
| #1of 50 | Gemini Omni 1.2 FlashGoogle | 1,503 | 2,666 | Gemini Omni 1.2 Flash |
| #4of 50 | Gemini Omni 1.1 FlashGoogle | 1,487 | 3,725 | Gemini Omni 1.1 Flash |
Arena's table of 9 Nov 2026.
The source
Arena's leaderboard, read from its public dataset on Hugging Face (CC BY 4.0), and refreshed by the daily job. Ratings, ranks and votes are Arena's; the matching to this site's roster is ours. See how the benchmarks fit the method.