Models are ranked by the quality of their English writing. By default the list shows the latest and previous model generations; tick “Older models” to add the rest. For the selected writing categories, ranks are calculated before publisher, size, access, reasoning and quantization filters; hidden models can therefore leave gaps.
Last updated:
| Rank | Model | AI judge score | Human score |
|---|---|---|---|
| 1 |  GPT-6.1 Sol 💡 | 95.7 | — |
| 2 |  Claude Opus 5.5 💡 | 94.8 | — |
| 3 |  Kimi K3 💡 | 92.2 | — |
| 4 | GPT-6 Astra 💡 | 91.9 | — |
| 5 | Claude Fable 5 💡 | 89.7 | — |
| 6 | Claude Fable 5.1 💡 | 89.4 | — |
| 7 | Claude Sonnet 5.5 💡 | 89.1 | — |
| 7 | GPT-6 Sol 💡 | 89.1 | — |
| 9 |  GLM 5.3 💡 | 88.7 | — |
| 10 | Claude Opus 5 💡 | 88.1 | — |
| 12 |  Grok 4.7 💡 | 87.6 | — |
| 13 | Claude Sonnet 5 💡 | 85.0 | — |
| 14 |  Aion 3.5 💡 | 84.0 | — |
| 15 | GLM 5.3 Flash 💡 | 81.4 | — |
| 16 | GPT-5.6 Sol 💡 | 80.5 | — |
| 17 |  MiMo V2.6 Pro 💡 | 79.7 | — |
| 18 |  MiniMax M3 💡 | 79.5 | — |
| 19 | GPT-6 Luna 💡 | 79.4 | — |
| 19 |  Muse Spark 1.3 💡 | 79.4 | — |
| 21 |  DeepSeek V4.1 Flash 💡 | 78.0 | — |
| 22 | Muse Spark 1.2 💡 | 77.7 | — |
| 23 | GLM 5.2 💡 | 77.6 | — |
| 24 |  Hy3 💡 | 77.3 | — |
| 25 |  Gemini 3.8 Flash 💡 | 76.9 | — |
| 26 | Claude Haiku 5.5 💡 | 74.1 | — |
| 27 |  Ling 3.1 Flash 💡 | 73.4 | — |
| 28 |  Qwen3.8 Max 💡 | 72.6 | — |
| 30 | Hy4 Preview 💡 | 70.3 | — |
| 32 | Aion 3.0 💡 | 69.6 | — |
| 33 | Gemini 3.7 Flash 💡 | 69.5 | — |
| 34 | Grok 4.6 💡 | 69.0 | — |
| 35 |  Mistral Large 4 💡 | 66.5 | — |
| 36 | GPT-5.6 Terra 💡 | 65.9 | — |
| 38 | GPT-5.6 Luna 💡 | 60.0 | — |
| 39 | DeepSeek V4 Pro 💡 | 59.3 | — |
| 41 | Qwen3.6 27B 💡 🏠 | 58.6 | — |
| 42 | Qwen3.7 Max 💡 | 57.7 | — |
| 43 | Qwen3.8 27B 💡 🏠 | 57.5 | — |
| 45 | DeepSeek V4 Flash 💡 | 55.2 | — |
| 47 | Gemma 4 31B 💡 🏠 | 53.5 | 58.5 |
| 48 | Qwen3.8 Flash 💡 | 52.9 | — |
| 49 |  Inkling 💡 | 52.2 | — |
| 53 | Qwen3.7 Plus 💡 | 50.2 | — |
| 54 | MiMo V2.6 Flash 💡 | 50.1 | — |
| 55 | Qwen3.6 35B A3B 💡 🏠 | 47.5 | 45.0 |
| 56 | MiniMax M3.1 Flash 💡 | 47.4 | — |
| 58 | MiMo V2.5 Pro 💡 | 45.3 | — |
| 59 | Gemma 4 31B ↯ 🏠 | 43.3 | — |
| 62 | MiMo V2.5 💡 | 42.1 | — |
| 65 | Gemma 4 12B 💡 🏠 | 41.3 | 55.0 |
| 65 | Gemma 4 26B A4B 💡 🏠 | 41.3 | 41.5 |
| 69 |  Nemotron 3 Ultra 💡 | 38.5 | — |
| 73 | Gemma 4 12B ↯ 🏠 | 34.9 | — |
| 74 |  Seed 2.1 Turbo ↯ | 34.2 | — |
| 75 | Gemma 4 26B A4B ↯ 🏠 | 32.0 | — |
| 76 | Nemotron 3.5 Lightning 💡 | 30.8 | — |
| 77 |  Agnes 3.0 Flash 💡 | 29.9 | — |
| 81 | Qwen3.6 27B ↯ 🏠 | 26.5 | — |
| 84 |  Granite 4.2 30B 💡 🏠 | 25.7 | — |
| 86 | Mistral Medium 3.5 💡 | 22.4 | — |
| 87 | Qwen3.6 35B A3B ↯ 🏠 | 21.2 | — |
| 91 | Qwen3.8 27B ↯ 🏠 | 19.3 | — |
| 92 | Granite 4.2 8B 💡 🏠 | 18.5 | — |
| 93 | Nemotron 3 Nano Omni 30B A3B 💡 🏠 | 18.4 | — |
| 95 | Gemma 4 E2B 💡 🏠 | 16.6 | — |
| 100 | Gemma 4 E4B 💡 🏠 | 13.4 | — |
| 101 | Gemma 4 E4B ↯ 🏠 | 11.8 | — |
| 105 | Gemma 4 E2B ↯ 🏠 | 10.4 | — |
A model appears after its first published AI-judged duel; its ranking score appears after its first eligible duel.
108 models compared115,791 published AI verdicts
Method: regularized Bradley–Terry model. The AI-judge score determines rank; the human score is independent and remains secondary until more community votes are available. A score of 50 means an estimated one-in-two chance of beating the average model in this language and category.
Each cell shows the share of available points earned by the row model against the column model: a win is worth 100%, a tie 50%, and a loss 0%. This result matrix uses only the compatible evaluation protocols included in the main ranking.
Directional matrix of points earned in direct duels. Values above 50% favour the row model; values below 50% favour the column model.
0 %100 %
💡 marks a model with reasoning, ↯ one without reasoning. 🏠 marks a model run locally on the site author’s computer.“on*”: the local file only has an on/off switch for reasoning, so no level can be chosen and the model runs at its default. “xhigh**”: Qwen3.8 27B runs at its documented default level, xhigh; high is not available on this model. For Muse Glimmer 30B and Ling 3.0 Flash (high) and for Granite 4.2 and Nemotron 3 Super (full), the level shown is the model’s documented default.
| Model |
|---|
Each cell shows how many published AI-judged duels across all published evaluation protocols compare the two models for this language and category. The models on both axes are candidates; this matrix does not show which judge produced each verdict.
Symmetric matrix of published duel counts. The diagonal is zero because a model is not compared with itself.
0
💡 marks a model with reasoning, ↯ one without reasoning. 🏠 marks a model run locally on the site author’s computer.“on*”: the local file only has an on/off switch for reasoning, so no level can be chosen and the model runs at its default. “xhigh**”: Qwen3.8 27B runs at its documented default level, xhigh; high is not available on this model. For Muse Glimmer 30B and Ling 3.0 Flash (high) and for Granite 4.2 and Nemotron 3 Super (full), the level shown is the model’s documented default.
| Model |
|---|
Which model writes the better text? Judge this anonymous pair and contribute to the human ranking. Model names stay hidden during the duel and are revealed only after your choice.
The leaderboard has one winner; your writing does not. On the duels of one kind of text alone, GPT-5.6 Sol takes fiction, Claude Opus 5 personal letters, Kimi K3 formal correspondence, and a local file you can run at home is second for cover letters. A plain guide, task by task.
We counted the marks readers call 'AI writing' in 3,350 ranked English texts: em dashes, stock phrases, 'not X but Y', lists of three. Kimi K3 and Claude Fable 5, the two best writers on 20 September, put an em dash every 80 words. The models with the fewest tells sit in the middle of the ranking, and the word 'delve' has all but vanished.
Ninety-one models are ranked in English, French, German and Spanish on the same protocol. Claude Fable 5 is third in all four; Claude Opus 5 first or second. Below the top, GLM 5.3 Flash is 10th in English and 54th in German, Grok 4.6 25th in English and 60th in French, and the best local model is a different model in every language.
The two most-asked-about assistants, judged on the same English prompts in blind duels: which one writes better, by how much, and for which kinds of text. Updated with every export of the ranking.
A live ranking of open-weight models for English writing, what each file needs in memory, and what fits in 16, 24, 32 or 128 GB.
