Google has introduced an upgraded preview of the Gemini 2.5 Pro model, set to become generally available soon. This update builds on the May release showcased at Google I/O, focusing on enterprise-scale applications. It showcases significant improvements, achieving a 24-point Elo increase on LMArena and a 35-point jump on WebDevArena, marking it as a leader in various benchmarks, including coding and academic performance.
Feedback from developers on previous iterations has been addressed, leading to enhancements in response quality and formatting. The upgraded model can be accessed via the Gemini API in Google AI Studio and Vertex AI, accompanied by new thinking budgets for cost and latency management. Additionally, Google unveiled Deep Think, which enhances reasoning capabilities, enabling the model to perform exceptionally well on the USA Mathematical Olympiad. Another innovation, Gemini Diffusion, was also announced, offering rapid output generation at 1,479 tokens/second.
Source link
