Writing and changing code. Split by how it is judged, because a grader running tests and a person picking the better of two answers do not measure the same thing and do not agree.
Each panel is one source's own ranking, shown as it publishes it. Nothing on this page is averaged across panels — where two disagree, both are right about what they measured.
Scored against tests or a reference, with no human in the loop and no agent loop either — these measure the model's output directly.
Coding Index fetched 2026-08-22
GPT-5.6 Sol 78.3 
Claude Opus 5 78.0 
Grok 4.6 76.8 GPT-5.6 Terra 76.7 Claude Fable 5 (Opus 4.8 fallback) 76.5 
Kimi K3 76.2 
Gemini 3.7 Flash 76.1 
GLM-5.3 74.8 Grok 4.5 72.4 
Muse Spark 1.2 72.2 SciCode fetched 2026-08-22
Claude Fable 5 (Opus 4.8 fallback) 60.2% Gemini 3.1 Pro Preview 58.9% Kimi K3 58.7% Gemini 3.7 Flash 57.9% GPT-5.6 Sol 56.9% GLM-5.3 56.5% Muse Spark 1.2 56.4% Claude Opus 5 55.7% Grok 4.6 54.6% Grok 4.5 54.1% Human preference between two built web applications. Arena's code arena is web work end to end — every category it publishes is a web application category — so this is the board it actually is, rather than a general coding board with a narrower name.
Overall as of 2026-08-21
Claude Opus 5 1691 Kimi K3 1674 Grok 4.6 1629 Claude Fable 5 (Opus 4.8 fallback) 1626 GPT-5.6 Sol 1619 GLM-5.3 1599 Gemini 3.7 Flash 1587 GLM-5.2 1582