0

SWE-bench Multilingual

Frontier

Cross-language extension of SWE-bench Verified - real GitHub issues across multiple programming languages.

Open
Publisher
Princeton University
Published
May 2026
Canonical
swebench.com/multilingual.html

Cite

Notes

Only stored in your browser.

Attribution

Leaderboard scores
SWE-bench
Attribution policy →

Top score 72.7% by Gemini 3 Flash - 11 models reporting (7 frontier)

Score history

11
30%48%65%83%100%Aug 25Oct 25Dec 25Feb 26GPT-5 MiniClaude Sonnet 4.5Gemini 3 ProClaude Opus 4.5Gemini 3 Flash

Top models

11
SWE-bench MultilingualBar chart with 11 bars. Highest value: Gemini 3 Flash at 72.7.
11 models

Related tools

2
View all

Implementations, trainers, datasets and scaffolds linked to this eval.

Agent Bench RL Env (Prime Community)

Prime Community

Benchmarking model performance on SWE Bench in the Mini SWE Agent harness.

Trains towardRL EnvTool UseAgent

SWE RL Env (Prime Intellect)

Prime Intellect

SWE tasks (R2E-Gym, SWE-bench, ...).

Trains towardRL EnvSWECode

FAQ

What is SWE-bench Multilingual?
Cross-language extension of SWE-bench Verified - real GitHub issues across multiple programming languages.
What is the current top score on SWE-bench Multilingual?
The top reported score is 72.7% by Gemini 3 Flash, across 11 models reporting (7 from frontier labs).
How can a model improve its SWE-bench Multilingual score?
Tools linked to SWE-bench Multilingual on Sophon include Agent Bench RL Env (Prime Community), SWE RL Env (Prime Intellect) - RL environments, datasets, and scaffolds that target this eval.