Coding Benchmarks

AI 模型代码能力评测排名

数据采集: 2026-04-16· Arena 数据: 2026-04-16

Arena Coding

Claude Opus 4.6 Thinking

1548

SWE-bench

live-SWE-agent + Claude 4.5 Opus medium (20251101)

79.2%

Aider

gpt-5 (high)

88.0%

LiveCodeBench

O4-Mini (High)

87.3%

Arms Race

各厂商最强模型的 Arena Coding Elo 趋势 · 每日更新

Arena Coding

基于用户投票的代码能力 Elo 排名 · arena.ai/leaderboard/webdev· 数据更新: 2026-04-16

1.Claude Opus 4.6 Thinking1548

2.Claude Opus 4.61545

3.GLM 5.11537

4.Claude Sonnet 4.61524

5.Claude Opus 4.5 20251101 Thinking 32k1490

6.Claude Opus 4.5 202511011468

7.GPT-5.4 High (codex Harness)1457

8.Gemini 3.1 Pro Preview1454

9.Qwen3.6 Plus Preview1453

10.GLM 51440

SWE-bench

真实 GitHub issue 修复能力 · swebench.com

1.live-SWE-agent + Claude 4.5 Opus medium (20251101)79.2%

2.Sonar Foundation Agent + Claude 4.5 Opus79.2%

3.TRAE + Doubao-Seed-Code78.8%

4.live-SWE-agent + Gemini 3 Pro Preview (2025-11-18)77.4%

5.Atlassian Rovo Dev (2025-09-02)76.8%

6.EPAM AI/Run Developer Agent v20250719 + Claude 4 Sonnet76.8%

7.mini-SWE-agent + Claude 4.5 Opus (high reasoning)76.8%

8.ACoder76.4%

9.mini-SWE-agent + Gemini 3 Flash (high reasoning)75.8%

10.mini-SWE-agent + MiniMax M2.5 (high reasoning)75.8%

Aider

代码编辑通过率 · aider.chat· 数据更新: 2026-03-17

1.gpt-5 (high)88.0%

2.gpt-5 (medium)86.7%

3.o3-pro (high)84.9%

4.gemini-2.5-pro-preview-06-05 (32k think)83.1%

5.o3 (high)81.3%

6.gpt-5 (low)81.3%

7.grok-4 (high)79.6%

8.gemini-2.5-pro-preview-06-05 (default think)79.1%

9.o3 (high) + gpt-4.178.2%

10.Gemini 2.5 Pro Preview 05-0676.9%

LiveCodeBench

竞赛编程（LeetCode / Codeforces）· livecodebench.github.io

1.O4-Mini (High)87.3%

2.O3 (High)84.7%

3.O4-Mini (Medium)84.5%

4.DeepSeek-R1-052884.4%

5.Gemini-2.5-Pro-06-0584.3%

6.Gemini-2.5-Pro-05-0682.7%

7.OpenReasoning-Nemotron-32B81.0%

8.EXAONE-4.0-32B80.9%

9.Qwen3-235B-A22B80.4%

10.XBai-o4-medium80.1%

每日自动更新 · Arena · SWE-bench · Aider · LiveCodeBench