benchmarks.site

Public leaderboard · Coding & software engineering

SWE-rebench

Tests AI coding tools on fresh real-world bug-fix tasks collected after models were trained, so they cannot have seen the answers.

Leaderboard

Top 3 right now

  1. 1stAnthropic Fable 5 [high]64.5
  2. 2ndGrok Grok 4.5 [high]63.8
  3. 3rdAnthropic Opus 5 [high]63.4

What it measures

Fresh real-world GitHub tasks collected in rolling time windows, scored by resolved rate and pass@5, with cost and tokens per problem.

How to read the numbers

Share of tasks resolved on the first attempt (%) — higher is better.

Results

17 rows · higher is better (Resolved Rate (%))CSV
1Anthropic Fable 5 [high]64.578.4$4.40
2Grok Grok 4.5 [high]63.877.5$1.47
3Anthropic Opus 5 [high]63.474.8$3.47
4Z.ai GLM-5.2 [high]62.981.1$1.40
5OpenAI GPT-5.6 Sol [medium]62.379.3$0.85
6Junie Junie61.873.9$0.81
7Anthropic Claude Code60.475.7$3.39
8OpenAI Codex58.073.0$1.59
9Anthropic Sonnet 5 [high]56.874.8$1.43
10Cursor Cursor51.765.8$0.41
11Minimax MiniMax M347.269.4$0.95
12XiaomiMiMo MiMo V2.5 Pro46.565.8$0.10
13OpenAI GPT-5.6 Luna [medium]43.659.5$0.11
14DeepSeek DeepSeek-V4 Pro [high]40.264.0$0.15
15Qwen Qwen3.6-27B31.257.7$0.62
16Qwen Qwen3.6-35B-A3B24.743.2$0.27
17Qwen Qwen3.5-35B-A3B17.136.9$0.99

Note: SWE-rebench v2, default time window 15/05/2026-01/07/2026 (111 problems). Entries without data in this window omitted. as_of is window end.

Source: SWE-rebench (Nebius) · swe-rebench.com · results as of 2026-07-01 · retrieved 2026-11-17

Markdown · More Coding & software engineering