Code Arena

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

Feb 9, 2026

143,756 votes

40 models

/

	Rank Spread
1	12	claude-opus-4-6-thinking Anthropic · Proprietary	1576+20/-20	1,268
2	12	claude-opus-4-6 Anthropic · Proprietary	1569+17/-17	1,824
3	33	claude-opus-4-5-20251101-thinking-32k Anthropic · Proprietary	1503+9/-9	9,409
4	47	gpt-5.2-high OpenAI · Proprietary	1472+16/-16	1,691
5	45	claude-opus-4-5-20251101 Anthropic · Proprietary	1471+8/-8	9,597
6	59	gemini-3-pro Google · Proprietary	1449+8/-8	15,589
7	510	kimi-k2.5-thinking Moonshot · Modified MIT	1446+12/-12	2,495
8	610	gemini-3-flash Google · Proprietary	1443+8/-8	11,158
9	610	glm-4.7 Z.ai · MIT	1441+10/-10	5,130
10	714	kimi-k2.5-instant Moonshot · Modified MIT	1421+17/-17	1,439
11	1016	gemini-3-flash (thinking-minimal) Google · Proprietary	1407+9/-9	7,208
12	1016	minimax-m2.1-preview MiniMax · MIT	1405+9/-9	8,466
13	1019	gpt-5.2 OpenAI · Proprietary	1397+16/-16	1,632
14	1019	gpt-5-medium OpenAI · Proprietary	1394+12/-12	3,926
15	1119	claude-opus-4-1-20250805 Anthropic · Proprietary	1389+8/-8	8,980
16	1119	gpt-5.1-medium OpenAI · Proprietary	1389+9/-9	6,438
17	1319	claude-sonnet-4-5-20250929-thinking-32k Anthropic · Proprietary	1388+7/-7	12,743
18	1319	claude-sonnet-4-5-20250929 Anthropic · Proprietary	1386+7/-7	14,337
19	1320	deepseek-v3.2-thinking DeepSeek · MIT	1375+10/-10	4,765
20	1922	glm-4.6 Z.ai · MIT	1357+8/-8	8,744
21	2023	gpt-5.1 OpenAI · Proprietary	1348+7/-7	11,643
22	2025	mimo-v2-flash (non-thinking) Xiaomi · MIT	1340+9/-9	5,535
23	2225	kimi-k2-thinking-turbo Moonshot · Modified MIT	1333+7/-7	11,143
24	2125	gpt-5.2-codex OpenAI · Proprietary	1333+10/-10	4,234
25	2226	gpt-5.1-codex OpenAI · Proprietary	1329+9/-9	6,502
26	2528	minimax-m2 MiniMax · Apache 2.0	1313+9/-9	8,833
27	2628	deepseek-v3.2 DeepSeek · MIT	1310+9/-9	6,028
28	2629	claude-haiku-4-5-20251001 Anthropic · Proprietary	1303+7/-7	12,415
29	2830	deepseek-v3.2-exp DeepSeek · MIT	1287+10/-10	5,131
30	2931	qwen3-coder-480b-a35b-instruct Alibaba · Apache 2.0	1280+7/-7	12,199
31	3032	KAT-Coder-Pro-V1 KwaiKAT · Proprietary	1259+15/-15	1,953
32	3134	gpt-5.1-codex-mini OpenAI · Proprietary	1243+17/-17	1,537
33	3234	grok-4-1-fast-reasoning xAI · Proprietary	1234+9/-9	6,886
34	3237	mistral-large-3 Mistral · Apache 2.0	1223+20/-20	1,037
35	3437	gemini-2.5-pro Google · Proprietary	1206+13/-13	3,453
36	3437	grok-4.1-thinking xAI · Proprietary	1204+19/-19	1,266
37	3437	devstral-2 Mistral · Modified MIT	1198+16/-16	1,678
38	3839	grok-4-fast-reasoning xAI · Proprietary	1153+22/-22	968
39	3840	grok-code-fast-1 xAI · Proprietary	1141+21/-21	1,016
40	3940	devstral-medium-2507 Mistral · Proprietary	1099+22/-22	1,021

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)

Battle Count for Each Combination of Models (without Ties)

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