← Back to Rankings
Model

AI Coding Capability Leaderboard

Data source: SWE-bench official Bash-Only leaderboard (mini-SWE-agent v2.0.0, 500 instances, single attempt). Data retrieved February 2026. LMSYS Chatbot Arena and other leaderboards temporarily unavailable due to network restrictions.

Rank Model Vendor SWE-bench Type
🥇 Claude 4.5 Opus Anthropic 76.8% Closed
🥈 Gemini 3 Flash Google DeepMind 75.8% Closed
🥉 MiniMax M2.5 MiniMax 75.8% Closed
4 Claude Opus 4.6 Anthropic 75.6% Closed
5 Claude 4.5 Opus (medium) Anthropic 74.4% Closed
6 Gemini 3 Pro Preview Google DeepMind 74.2% Closed
7 GLM-5 Z-AI 72.8% Closed
8 GPT-5.2 OpenAI 72.8% Closed
9 Claude 4.5 Sonnet Anthropic 71.4% Closed
10 Kimi K2.5 Moonshot AI 70.8% Closed
11 DeepSeek V3.2 DeepSeek 70.0% Open
12 Gemini 3 Pro Google DeepMind 69.6% Closed
13 Claude 4 Opus Anthropic 67.6% Closed
14 Claude 4.5 Haiku Anthropic 66.6% Closed
15 GPT-5.1 OpenAI 66.0% Closed
16 GPT-5 OpenAI 65.0% Closed
17 Claude 4 Sonnet Anthropic 64.9% Closed