Google introduces upgraded Gemini 2.5 Pro preview with improved coding and reasoning

Google has released an upgraded preview of its Gemini 2.5 Pro model, positioning it as its most intelligent offering to date ahead of a general availability launch in a few weeks.

• The model achieved a 24-point Elo score jump on LMArena (1470) and a 35-point jump on WebDevArena (1443).
• It maintains leadership on difficult coding benchmarks like Aider Polyglot while showing top-tier performance on GPQA and Humanity’s Last Exam.
• The update addresses previous feedback by improving response style, structure, and creativity.
• Developers can access the model via Gemini API in Google AI Studio and Vertex AI, which now include thinking budgets for cost and latency control.

The upgraded preview is available today in the Gemini app and provides developers with enhanced control over model behavior through new budgeting features.