Code Arena | React

View overall rankings across AI models on agentic coding tasks involving multi-step reasoning, tool use, and production-style workflows.

Feb 24, 2026

47,939 votes

29 models

	Rank Spread
1	13	claude-opus-4-6 Anthropic · Proprietary	1547+14/-14	2,424
2	13	claude-opus-4-6-thinking Anthropic · Proprietary	1539+16/-16	1,815
3	13	claude-sonnet-4-6 Anthropic · Proprietary	1521+17/-17	1,517
4	46	claude-opus-4-5-20251101-thinking-32k Anthropic · Proprietary	1485+12/-12	3,057
5	46	claude-opus-4-5-20251101 Anthropic · Proprietary	1470+12/-12	2,968
6	611	glm-5 Z.ai · MIT	1438+14/-14	2,271
7	611	gemini-3.1-pro-preview Google · Proprietary	1437+16/-16	1,604
8	611	gemini-3-flash Google · Proprietary	1432+11/-11	3,361
9	419	glm-4.7 Z.ai · MIT	1426+57/-57	128
10	612	kimi-k2.5-thinking Moonshot · Modified MIT	1425+11/-11	3,272
11	613	minimax-m2.5 MiniMax · Modified MIT	1420+12/-12	3,122
12	917	kimi-k2.5-instant Moonshot · Modified MIT	1404+13/-13	2,455
13	1018	gemini-3-pro Google · Proprietary	1400+12/-12	2,745
14	1118	claude-sonnet-4-5-20250929 Anthropic · Proprietary	1386+12/-12	2,828
15	1119	qwen3.5-397b-a17b Alibaba · Apache 2.0	1384+13/-13	2,142
16	1119	claude-sonnet-4-5-20250929-thinking-32k Anthropic · Proprietary	1383+12/-12	2,671
17	1119	gemini-3-flash (thinking-minimal) Google · Proprietary	1380+11/-11	3,108
18	1219	minimax-m2.1-preview MiniMax · MIT	1377+13/-13	2,496
19	1420	deepseek-v3.2-thinking DeepSeek · MIT	1359+13/-13	2,188
20	1921	deepseek-v3.2 DeepSeek · MIT	1342+13/-13	2,354
21	2025	gpt-5.2-codex OpenAI · Proprietary	1323+12/-12	2,821
22	2125	mimo-v2-flash (non-thinking) Xiaomi · MIT	1314+12/-12	2,599
23	2125	claude-haiku-4-5-20251001 Anthropic · Proprietary	1311+12/-12	2,707
24	2125	gpt-5.1 OpenAI · Proprietary	1310+12/-12	2,735
25	2125	kimi-k2-thinking-turbo Moonshot · Modified MIT	1309+12/-12	2,631
26	2627	qwen3-coder-480b-a35b-instruct Alibaba · Apache 2.0	1256+13/-13	2,671
27	2628	mimo-v2-flash (thinking) Xiaomi · MIT	1237+20/-20	928
28	2728	grok-4-1-fast-reasoning xAI · Proprietary	1225+17/-17	1,468
29	2929	devstral-2 Mistral · Modified MIT	1159+42/-42	231

Code Arena | React

Remove Style Control Leaderboard Plots

Confidence Intervals on Model Strength (via Bootstrapping)

Battle Count for Each Combination of Models (without Ties)

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

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