Google has recently witnessed a significant update in its series of AI models. Following the launch of the Gemini2.5Pro model at the end of March this year, Google promptly released the more lightweight Gemini2.5Flash version in April. Although both are currently in preview stages, the Gemini2.5Flash version is already available to global users through the Gemini App, allowing them to experience its powerful features.
Then, on June 6th, Google once again updated the Gemini2.5Pro model to version 06-05. The focus of this update was a substantial performance improvement. Google claims that the model achieved higher scores in multiple AI performance benchmark tests, particularly excelling in tests like LMArena and WebDevArena, where the Elo ratings increased by 24 points and 35 points respectively. The latter even reached 1443 points, becoming a leader in the industry.
In some challenging tests such as GPQA and "The Final Exam for Humans," Gemini2.5Pro also performed excellently, showcasing its strong intelligent capabilities. Additionally, Google responded to user feedback about performance fluctuations in previous versions, stating that it has optimized the model in terms of style and structure, making its responses more creative and clearer, thereby enhancing user interaction experiences.
It's worth noting that Google has relaxed request restrictions for Pro version users, meaning users will enjoy greater flexibility and convenience when using the model. This change may foreshadow more attractive pricing strategies from Google after the official release of this version later this month, further encouraging user adoption.
This update to Gemini2.5Pro not only enhances AI performance and user experience but also signals Google's ongoing innovation and improvement in the AI field.