When you purchase through links on our site, we may earn an affiliate commission. Here’s how it works. Google announces Gemini 2.5 Pro Deep Think, beating OpenAI's o3 and o4 models Pradeep Viswanathan Neowin @pradeepviswav · May 20, 2025 16:32 EDT At Google I/O 2025, Google announced today several updates to the Gemini 2.5 model series. The main highlight was the Gemini 2.5 Pro Deep Think mode, which reportedly beats OpenAI's latest o3 and o4 series models in popular AI benchmarks. Google didn't announce any update to the Gemini 2.5 Pro model since it received a significant update just earlier this month. However, the new enhanced reasoning mode, called Deep Think, will push the capabilities of the 2.5 Pro model to the next level. Deep Think will use new research techniques to consider multiple hypotheses before responding. Google shared the following three benchmarks for 2.5 Pro Deep Think: 49.4% on the 2025 USAMO math benchmarks. 80.4% on the LiveCodeBench competition-level coding benchmark. 84.0% on the MMMU multimodal reasoning benchmark. All the above scores are new SOTA, beating even the latest OpenAI's o3 and o4 series models. 2.5 Pro Deep Think will be available to trusted testers via the Gemini API for now. Google also unveiled the latest update to Gemini 2.5 Flash, its low-cost model. This new model performs better than the previous version in every benchmark, and it is now available for preview in Google AI Studio for developers, in Vertex AI for enterprise, and in the Gemini app. Google will release the production version of 2.5 Flash in June. Along with the model updates, Google announced the following improvements to the Gemini developer experience: A new Live API preview with support for multiple speakers, enabling text-to-speech with two voices via native audio output, and more. Native SDK support for Model Context Protocol (MCP) definitions in the Gemini API for easier integration with open-source tools. Gemini 2.5 Pro with thinking budgets will be generally available for stable production use in the coming weeks. Project Mariner's computer use capabilities are coming to the Gemini API and Vertex AI. 2.5 Pro and Flash will now include thought summaries in the Gemini API and in Vertex AI. You can learn more about Gemini 2.5 model updates here. Tags Report a problem with article Follow @NeowinFeed