Leaderboard

Rank	Diff	Model	Score	95% CI	Votes	License
1	—	grok-4-fast xAI	1182		32	Proprietary
2	—	gemini-3-pro-preview Google	1102		28	Proprietary
3	—	gemini-3.1-pro-preview Google	1078		12	Proprietary
4	—	deepseek-v3.1-terminus DeepSeek	1074		39	Open Source
5	—	claude-sonnet-4.6 Anthropic	1063		21	Proprietary
6	—	claude-opus-4.1 Anthropic	1046		34	Proprietary
7	—	grok-4.1-fast xAI	1035		28	Proprietary
8	—	gpt-4.1 OpenAI	1026		35	Proprietary
9	—	qwen3-vl-235b-a22b-instruct Qwen	1003		24	Open Source
10	—	gpt-5.1 OpenAI	1002		25	Proprietary
11	—	grok-4 xAI	995		33	Proprietary
12	—	kimi-k2-0905 Moonshot AI	967		35	Open Source
13	—	gpt-5.2 OpenAI	950		22	Proprietary
14	—	gpt-5-chat OpenAI	940		35	Proprietary
15	—	o3 OpenAI	926		24	Proprietary
16	—	glm-4.6 Zhipu AI	922		26	Open Source
17	—	deepseek-r1-0528 DeepSeek	916		37	Open Source
18	▲ 1	deepseek-chat-v3.1 DeepSeek	905		23	Open Source
19	▲ 1	qwen3-235b-a22b-2507 Qwen	900		30	Open Source
20	▼ 2	qwen3-max Qwen	900		38	Proprietary
21	—	claude-opus-4 Anthropic	887		34	Proprietary
22	▲ 5	kimi-k2 Moonshot AI	854		27	Open Source
23	▼ 1	gemini-2.5-pro Google	850		31	Proprietary
24	▼ 1	claude-sonnet-4.5 Anthropic	848		37	Proprietary
25	▼ 1	gpt-4o OpenAI	845		12	Proprietary

Models are ranked by Bradley-Terry score on an Elo scale. 95% bootstrap confidence intervals show statistical uncertainty; once significance-aware ranking is enabled, models with overlapping intervals will share a tied rank.