SWE-bench Multilingual

Resolving real GitHub issues across 300 tasks in 9+ programming languages, scored by whether the patch passes the repo tests. Source: swebench.com multilingual leaderboard (Feb 2026).

Features measured

Code Tool use Reasoning Long context Multilingual

Use cases unlocked

Autonomous coding agents Bug fixing Coding & code review

Leaderboard

Higher is better · % resolved

1 Gemini 3 Flash 72.7%

2 Claude 4.6 Opus 72%

3 Claude 4.5 Opus 70.7%

4 GLM 5 69.7%

5 Gemini 3 Pro 68.7%

6 MiniMax M2.5 68.3%

7 Kimi K2.5 67.3%

8 Claude 4.5 Sonnet 67%

Bars scaled from 60 so close scores stay legible — the figure at right is the actual % resolved.

Benchmark source ↗

← All benchmarks