Code Arena

Compare the performance of AI models on agentic coding tasks involving multi-step reasoning and tool use

Last Updated

Feb 6, 2026

Total Votes

136,159

Total Models

/

	Rank Spread
1	1◄─►1	claude-opus-4-6	1576	+19/-19	1,422	Anthropic	Proprietary
2	2◄─►2	claude-opus-4-5-20251101-thinking-32k	1502	+9/-9	9,003	Anthropic	Proprietary
3	3◄─►6	gpt-5.2-high	1472	+16/-16	1,691	OpenAI	Proprietary
4	3◄─►5	claude-opus-4-5-20251101	1470	+9/-9	9,179	Anthropic	Proprietary
5	4◄─►8	gemini-3-pro	1452	+8/-8	15,193	Google	Proprietary
6	3◄─►8	kimi-k2.5-thinking	1449	+14/-14	2,123	Moonshot	Modified MIT
7	5◄─►8	gemini-3-flash	1442	+8/-8	10,736	Google	Proprietary
8	5◄─►8	glm-4.7	1441	+10/-10	5,125	Z.ai	MIT
9	9◄─►13	minimax-m2.1-preview	1408	+9/-9	8,095	MiniMax	MIT
10	9◄─►17	kimi-k2.5-instant	1407	+19/-19	1,056	Moonshot	Modified MIT
11	9◄─►15	gemini-3-flash (thinking-minimal)	1406	+9/-9	6,788	Google	Proprietary
12	9◄─►18	gpt-5.2	1397	+16/-16	1,632	OpenAI	Proprietary
13	9◄─►18	gpt-5-medium	1394	+12/-12	3,925	OpenAI	Proprietary
14	10◄─►18	claude-opus-4-1-20250805	1389	+9/-9	8,980	Anthropic	Proprietary
15	10◄─►18	gpt-5.1-medium	1389	+9/-9	6,432	OpenAI	Proprietary
16	11◄─►18	claude-sonnet-4-5-20250929-thinking-32k	1387	+7/-7	12,309	Anthropic	Proprietary
17	11◄─►18	claude-sonnet-4-5-20250929	1386	+7/-7	13,951	Anthropic	Proprietary
18	12◄─►19	deepseek-v3.2-thinking	1374	+10/-10	4,449	DeepSeek	MIT
19	18◄─►21	glm-4.6	1357	+9/-9	8,741	Z.ai	MIT
20	19◄─►22	gpt-5.1	1349	+8/-8	11,221	OpenAI	Proprietary
21	19◄─►24	mimo-v2-flash (non-thinking)	1344	+9/-9	5,156	Xiaomi	MIT
22	20◄─►24	gpt-5.2-codex	1336	+11/-11	3,852	OpenAI	Proprietary
23	21◄─►24	kimi-k2-thinking-turbo	1331	+8/-8	10,780	Moonshot	Modified MIT
24	21◄─►25	gpt-5.1-codex	1329	+9/-9	6,501	OpenAI	Proprietary
25	24◄─►27	minimax-m2	1313	+9/-9	8,833	MiniMax	Apache 2.0
26	25◄─►27	deepseek-v3.2	1309	+9/-9	5,654	DeepSeek	MIT
27	25◄─►28	claude-haiku-4-5-20251001	1301	+7/-7	12,024	Anthropic	Proprietary
28	27◄─►29	deepseek-v3.2-exp	1287	+10/-10	5,130	DeepSeek	MIT
29	28◄─►30	qwen3-coder-480b-a35b-instruct	1281	+7/-7	11,785	Alibaba	Apache 2.0
30	29◄─►32	KAT-Coder-Pro-V1	1259	+15/-15	1,954	KwaiKAT	Proprietary
31	30◄─►33	gpt-5.1-codex-mini	1243	+17/-17	1,537	OpenAI	Proprietary
32	30◄─►33	grok-4-1-fast-reasoning	1235	+10/-10	6,480	xAI	Proprietary
33	31◄─►36	mistral-large-3	1223	+20/-20	1,037	Mistral	Apache 2.0
34	33◄─►36	gemini-2.5-pro	1206	+13/-13	3,454	Google	Proprietary
35	33◄─►36	grok-4.1-thinking	1205	+19/-19	1,265	xAI	Proprietary
36	33◄─►36	devstral-2	1199	+16/-16	1,678	Mistral	Modified MIT
37	37◄─►38	grok-4-fast-reasoning	1153	+23/-23	968	xAI	Proprietary
38	37◄─►39	grok-code-fast-1	1141	+21/-21	1,016	xAI	Proprietary
39	38◄─►39	devstral-medium-2507	1099	+22/-22	1,021	Mistral	Proprietary

Code Arena

Remove Style Control Leaderboard Plots

Fraction of Model A Wins for All Non-tied A vs. B Battles

Confidence Intervals on Model Strength (via Bootstrapping)

Average Win Rate Against All Other Models (Uniform Sampling and No Ties)

Battle Count for Each Combination of Models (without Ties)