Google has released the Gemini 2.5 06-05 preview, introducing significant improvements for developers in AI-driven coding, reasoning, and multimodal tasks. This release aims to position Gemini as a leading option for technical teams building, testing, and scaling AI-powered applications.
- Achieves an 82.2% score on the Aider Polyglot benchmark, outperforming competitors on real-world coding tasks.
- Reaches a 1470 Elo rating on LMArena and 1443 on WebDev Arena, marking substantial gains in logic and web app generation.
- Supports a 1 million token context window, capable of processing large codebases or extended media in a single call.
- Demonstrates top-tier performance on the VideoMME benchmark with an 84.8% score for video understanding.
- Offers configurable thinking budgets to allow developers to fine-tune the tradeoff between cost and latency.
The update is available now via Google AI Studio, Vertex AI, and the Gemini app, with general availability planned for stable production use.