Models
Every model in the arena, ranked by blind human judgment.
All models
GPT-6.1 Sol
1–2 · 1145 ±30
Muse Wonder 1.0
1–3 · 1110 ±48
Claude Opus 5.5
2–5 · 1064 ±28
Claude Fable 5
3–6 · 1047 ±22
Gemini 4 Argon
3–8 · 1042 ±30
GPT-6 Astra
4–11 · 1023 ±31
Claude Fable 5.1
5–11 · 1017 ±24
GPT-6 Luna
5–14 · 1006 ±29
Kimi K3
6–14 · 1002 ±23
Gemini 3.8 Flash
6–14 · 996 ±30
Qwen3.8-2.4T-A95B
7–15 · 994 ±28
Grok 4.7
7–15 · 991 ±26
Claude Sonnet 5.5
7–15 · 990 ±31
Claude Haiku 5.5
10–16 · 969 ±30
Muse Spark 1.3
7–17 · 966 ±57
MiMo V2.6 Pro
15–17 · 931 ±32
Inkling
15–18 · 924 ±24
GLM 5.3 Flash
17–18 · 888 ±28
Rank band, then rating and its 95% interval, drawn on one scale with lines at 900, 1000 and 1100. n = 18 models, each rated from at least 30 comparisons. Last updated .