Introduction to Gemini 2.5 Pro
We are excited to announce an enhanced preview of Gemini 2.5 Pro, our most advanced model to date. This version improves upon the model we released in May and showcased at I/O. In a few weeks, this model will become the standard, stable version, ready for large-scale enterprise applications.

Performance Updates
The latest Gemini 2.5 Pro has demonstrated significant improvements, with a 24-point Elo score increase on LMArena, solidifying its lead on the leaderboard at 1470. Additionally, it has achieved a 35-point Elo jump, taking the lead on WebDevArena at 1443. Gemini 2.5 Pro continues to excel in coding, outperforming others on challenging coding benchmarks like Aider Polyglot. It also showcases top-tier performance on GPQA and Humanity’s Last Exam (HLE), highly demanding benchmarks that assess a model’s mathematical, scientific, knowledge, and reasoning abilities.

Enhancements and Improvements
We have addressed feedback from our previous 2.5 Pro release, refining its style and structure to produce more creative and better-formatted responses.

Availability and Access
Developers can now access the upgraded preview of Gemini 2.5 Pro through the Gemini API on Google AI Studio and Vertex AI. We have also introduced thinking budgets, providing developers with greater control over cost and latency. Furthermore, the updated model is rolling out today in the Gemini app.