SWE-rebench Benchmark Releases Multilingual Update for Software Engineering Tasks

The SWE-rebench benchmark has released a major multilingual update, expanding its evaluation of software engineering tasks beyond Python to include Go, Java, Rust, and TypeScript. The update evaluates several open-weight models, including GLM-5.2, DeepSeek-V4 Pro, and various Qwen models. Evaluating LLMs on multiple programming languages provides a more realistic measure of their utility in diverse, real-world software development environments. It helps developers identify which models perform best for specific languages and local development setups. In the latest evaluation, GLM-5.2 achieved the highest performance with a 62.9% Pass@1 rate and 81.1% Pass@5 rate, successfully passing tasks in all five languages 39.6% of the time. The benchmark organizers plan to release another update in 3-4 weeks focusing on models optimized for local deployment.