Google Unveils Deep Think to Enhance Gemini AI Performance
Google is advancing its top-tier Gemini AI models.
At Google I/O 2025 on Tuesday, the company introduced Deep Think, an advanced reasoning mode for its premier Gemini 2.5 Pro model. Deep Think enables the model to evaluate multiple responses before answering, improving its results on specific benchmarks.
“Deep Think maximizes model capabilities,” said Demis Hassabis, leader of Google DeepMind, the company’s AI research division, during a press briefing. “It leverages our most advanced research in reasoning and parallel processing techniques.”
Google remained discreet about Deep Think’s mechanics, but the technology may resemble OpenAI’s o1-pro and forthcoming o3-pro models, which likely employ an engine to identify and refine optimal solutions for complex queries.
Google reports that Deep Think propelled Gemini 2.5 Pro to lead LiveCodeBench, a rigorous coding assessment. Additionally, Gemini 2.5 Pro with Deep Think surpassed OpenAI’s o3 on MMMU, a benchmark testing skills like perception and reasoning.

Image Credits: Google DeepMind Deep Think is accessible to “trusted testers” through the Gemini API starting this week. Google noted it is conducting thorough safety assessments before a broader rollout.
In addition to Deep Think, Google unveiled an upgrade to its cost-effective Gemini 2.5 Flash model, enhancing its performance in coding, multimodality, reasoning, and handling extended contexts. The updated, more efficient Gemini 2.5 Flash is available for preview in Google’s AI Studio, Vertex AI platforms, and Gemini apps.
Join us at TechCrunch Sessions: AI
Reserve your place at our premier AI industry event featuring speakers from OpenAI, Anthropic, and Cohere. For a limited time, tickets are only $292 for a full day of expert talks, workshops, and impactful networking.
Exhibit at TechCrunch Sessions: AI
Secure your spot at TC Sessions: AI to showcase your innovations to over 1,200 decision-makers—without breaking the bank. Available through May 9 or while tables last.
Berkeley, CA | June 5 REGISTER NOWGoogle stated that the enhanced Gemini 2.5 Flash will be broadly available to developers in June.
Finally, Google introduced Gemini Diffusion, a model touted as “highly efficient,” delivering outputs four to five times faster than similar models while matching the performance of those twice its size. Gemini Diffusion is available to “trusted testers” starting today.
Related article
Google integrates agentic AI and vibe-coded widgets into Android
Google announced a fresh batch of AI features under its Gemini Intelligence brand during the “Android Show: I/O Edition” event on Tuesday. These capabilities include having the AI handle tasks across multiple apps, browse the web, fill out forms, tra
Meta's AI model excels but open-source identity erodes
The open-source AI landscape has always offered plenty of choices. For years, developers could access models like Mistral, Falcon, and a growing number of open-weight alternatives. But Meta's entry with Llama changed the game. A company with three bi
Father sues Google, blames Gemini chatbot for son's fatal delusion
Jonathan Gavalas, 36, began using Google's Gemini AI chatbot in August 2025 for shopping assistance, writing help, and travel planning. On October 2, he died by suicide. At the time of his death, he believed Gemini was his fully sentient AI wife and
Related Special Topic Recommendations
Comments (3)
0/500
Гугл снова удивляет! Deep Think звучит как серьёзный шаг в AI-рассуждениях, но не приведёт ли это к ещё большей зависимости от одной корпорации? 🤔 Интересно, как это повлияет на маленькие стартапы в сфере ИИ.
Tja, 'Deep Think' hört sich nach noch mehr Werbe-Hype an 🙄 Google kommt einfach nicht von seinem Marketing-Bullshit-Bingo los. Was heißt 'fortgeschrittener Verarbeitungsmodus' konkret? Statt neue Buzzwords zu erfinden, sollten sie lieber ihre Probleme mit halbgaren KI-Antworten fixen. Die Demos auf I/O sahen immer toll aus, aber wer weiß, wie es im Alltag performt. Ehrlich gesagt glaube ich erst daran, wenn es für den normalen User wirklich besser läuft.
Google is advancing its top-tier Gemini AI models.
At Google I/O 2025 on Tuesday, the company introduced Deep Think, an advanced reasoning mode for its premier Gemini 2.5 Pro model. Deep Think enables the model to evaluate multiple responses before answering, improving its results on specific benchmarks.
“Deep Think maximizes model capabilities,” said Demis Hassabis, leader of Google DeepMind, the company’s AI research division, during a press briefing. “It leverages our most advanced research in reasoning and parallel processing techniques.”
Google remained discreet about Deep Think’s mechanics, but the technology may resemble OpenAI’s o1-pro and forthcoming o3-pro models, which likely employ an engine to identify and refine optimal solutions for complex queries.
Google reports that Deep Think propelled Gemini 2.5 Pro to lead LiveCodeBench, a rigorous coding assessment. Additionally, Gemini 2.5 Pro with Deep Think surpassed OpenAI’s o3 on MMMU, a benchmark testing skills like perception and reasoning.

Deep Think is accessible to “trusted testers” through the Gemini API starting this week. Google noted it is conducting thorough safety assessments before a broader rollout.
In addition to Deep Think, Google unveiled an upgrade to its cost-effective Gemini 2.5 Flash model, enhancing its performance in coding, multimodality, reasoning, and handling extended contexts. The updated, more efficient Gemini 2.5 Flash is available for preview in Google’s AI Studio, Vertex AI platforms, and Gemini apps.
Join us at TechCrunch Sessions: AI
Reserve your place at our premier AI industry event featuring speakers from OpenAI, Anthropic, and Cohere. For a limited time, tickets are only $292 for a full day of expert talks, workshops, and impactful networking.
Exhibit at TechCrunch Sessions: AI
Secure your spot at TC Sessions: AI to showcase your innovations to over 1,200 decision-makers—without breaking the bank. Available through May 9 or while tables last.
Berkeley, CA | June 5 REGISTER NOWGoogle stated that the enhanced Gemini 2.5 Flash will be broadly available to developers in June.
Finally, Google introduced Gemini Diffusion, a model touted as “highly efficient,” delivering outputs four to five times faster than similar models while matching the performance of those twice its size. Gemini Diffusion is available to “trusted testers” starting today.
Google integrates agentic AI and vibe-coded widgets into Android
Google announced a fresh batch of AI features under its Gemini Intelligence brand during the “Android Show: I/O Edition” event on Tuesday. These capabilities include having the AI handle tasks across multiple apps, browse the web, fill out forms, tra
Meta's AI model excels but open-source identity erodes
The open-source AI landscape has always offered plenty of choices. For years, developers could access models like Mistral, Falcon, and a growing number of open-weight alternatives. But Meta's entry with Llama changed the game. A company with three bi
Father sues Google, blames Gemini chatbot for son's fatal delusion
Jonathan Gavalas, 36, began using Google's Gemini AI chatbot in August 2025 for shopping assistance, writing help, and travel planning. On October 2, he died by suicide. At the time of his death, he believed Gemini was his fully sentient AI wife and
Гугл снова удивляет! Deep Think звучит как серьёзный шаг в AI-рассуждениях, но не приведёт ли это к ещё большей зависимости от одной корпорации? 🤔 Интересно, как это повлияет на маленькие стартапы в сфере ИИ.
Tja, 'Deep Think' hört sich nach noch mehr Werbe-Hype an 🙄 Google kommt einfach nicht von seinem Marketing-Bullshit-Bingo los. Was heißt 'fortgeschrittener Verarbeitungsmodus' konkret? Statt neue Buzzwords zu erfinden, sollten sie lieber ihre Probleme mit halbgaren KI-Antworten fixen. Die Demos auf I/O sahen immer toll aus, aber wer weiß, wie es im Alltag performt. Ehrlich gesagt glaube ich erst daran, wenn es für den normalen User wirklich besser läuft.





Home






