Google Gemini 2.5 Pro Achieves State-of-the-Art on Agentic Coding Benchmarks

Google has released Gemini 2.5 Pro, which achieves new state-of-the-art results on multiple agentic coding benchmarks including SWE-Bench, HumanEval, and the newly introduced CodeArena evaluation suite. The model successfully resolves 72 percent of real-world GitHub issues in the SWE-Bench test.

The model features a two-million-token context window and an improved architecture that enables it to maintain coherent reasoning across large codebases. Google reports that Gemini 2.5 Pro can autonomously refactor entire software projects while preserving existing test coverage.

Enterprise customers can access the model through Google Cloud's Vertex AI platform, with integration into popular development environments including VS Code and JetBrains IDEs.