TipsMake
Newest
• Czkawka - An open-source application that helps remove duplicate data. • ESM bio-concrete: The first building material capable of absorbing carbon emissions. • 6 types of physical vehicles that are far better than their digital versions. • Sony returns to the vinyl market with two brand-new Bluetooth turntables. • Is Gigabit Ethernet still sufficient for the needs of modern home networks?

Google Announces Gemini 2.5 Pro Deep Think, Beats OpenAI's o3 and o4 Models

★ What is Google Gemini? How does Gemini work?
★ Instructions for using the Gemini utility

At Google I/O 2025, Google announced a number of updates to its Gemini 2.5 model line. The main highlight was the Gemini 2.5 Pro Deep Think mode, which is said to beat OpenAI's latest o3 and o4 series models in popular AI benchmarks.

 

Google isn't announcing any updates to the Gemini 2.5 Pro model , which just received a major update earlier this month. However, a new advanced inference mode, called Deep Think, will take the 2.5 Pro model's capabilities to the next level. Deep Think will use new research techniques to consider multiple hypotheses before responding.

Google has shared the following 3 benchmarks for the 2.5 Pro Deep Think:

  1. 49.4% in the USAMO 2025 math benchmark.
  2. 80.4% in the LiveCodeBench competitive-level coding benchmark.
  3. 84.0% in the MMMU multimodal inference benchmark.

 

All of the above scores are new SOTA, even beating OpenAI's latest o3 and o4 series models. 2.5 Pro Deep Think will be available to trusted testers via the Gemini API right now.

Google also announced a new update to Gemini 2.5 Flash, its low-cost model. The new model outperforms the previous version in every benchmark and is available for preview in Google AI Studio for developers, Vertex AI for enterprises, and the Gemini app. Google will release a production version of 2.5 Flash in June.

Along with the model updates, Google announced the following improvements to the Gemini developer experience:

  1. The new live API preview supports multiple speakers, enables dual-voice text-to-speech via native audio output, etc.
  2. The native SDK supports Model Context Protocol (MCP) definitions in the Gemini API for easier integration with open source tools.
  3. Gemini 2.5 Pro with 'Thinking budget' (a concept that refers to a budgeting approach based on deep thinking, thorough analysis, and decision making based on available data and information) will be available for stable production use in the coming weeks.
  4. Project Mariner's computing capabilities will be available in the Gemini API and Vertex AI.
    2.5 Pro. Flash will now include thought summaries in the Gemini API and in Vertex AI.

You can learn more about the Gemini 2.5 model updates here .

Discover more
Micah Soto
Share by Micah Soto
Update 26 May 2025

You should read it