Google DeepMind has launched Gemini 2.5 Pro, a new version of its AI model built to handle complex reasoning, answer long-form questions, and write code. The model is now available to Gemini Advanced users in Google AI Studio, with plans to roll it out on Vertex AI, Google’s Machine Learning AI platform in the immediate future.
Gemini 2.5 Pro scored among the highest on several technical tests, including the AIME math exam and the GPQA science benchmark. It also ranked first on LMArena, a live, community-driven evaluation where anonymous LLMs are compared side by side and rated by the human evaluators.
In software tasks, the model scored 63.8% on SWE-Bench Verified, a benchmark that checks if an AI can solve real issues from GitHub. Google says Gemini 2.5 Pro can now build working web apps from a single prompt and do more advanced code changes than earlier versions.
Behind the scenes, Google claims the model is also faster and more memory-efficient. Google says this will help reduce wait times for responses as the model is used in apps and cloud services.
While it will take time to get the full reaction of users in real world settings, the early indications is that Gemini 2.5 puts Google back in serious contention with the other leading frontier models.
