Google’s new Gemini 2.5 Pro release aims to fix past “regressions” in the model

Google has released an updated version of its Gemini Pro AI model, aiming to address previous issues and solidify its position as a reliable tool for coding tasks. The new 06-05 update boasts improved performance, scoring 82.2% in the Aider Polyglot test, outperforming models from OpenAI, Anthropic, and DeepSeek. Logan Kilpatrick highlighted feedback improvements that address shortcomings seen since an earlier release, with enhancements to creativity and response formatting. The team has also introduced configurable thinking budgets for developers. This version marks a step toward long-term stability as Google finalizes its preview phase.