Google Releases Advanced AI Model for Complex Reasoning Tasks

Google has released Gemini 2.5 Deep Think, an advanced artificial intelligence model designed for complex reasoning tasks.

The model uses extended processing time to analyze multiple approaches to problems simultaneously before generating responses, Google said. Deep Think takes several minutes to produce answers compared to standard AI models that respond in seconds.

Google first previewed the technology at its I/O conference in May and has been testing it with select users. It's now available to subscribers of the company's $250-per-month AI Ultra plan. Google said the public version achieves bronze-level performance on International Mathematical Olympiad problems, while a research variant that takes hours to process queries recently earned gold-medal status.

The model scored 87.6% on LiveCodeBench, a competitive coding benchmark, compared to 80.4% in May testing. On Humanity's Last Exam, a multi-subject assessment covering advanced topics, Deep Think achieved 34.8% accuracy versus 20-25% for competing models.

AI Ultra subscribers receive limited daily queries with Deep Think through Google's Gemini interface, though the company did not specify exact usage limits. The service includes access to experimental features, increased storage, and prototype tools.

Google plans to offer Deep Think through its developer API in the coming weeks for enterprise and research applications. The model competes with OpenAI's o1 and o3 reasoning models, which use similar extended processing approaches.

The release highlights the growing competition among technology companies to develop AI systems capable of managing complex mathematical, scientific, and coding tasks. Google's premium pricing restricts access compared to standard AI models that are available for free or through lower-cost subscriptions.

Google has not disclosed adoption figures for AI Ultra or revenue from its AI services. "Because we're defining the frontier with 2.5 Pro DeepThink, we're taking extra time to conduct more frontier safety evaluations and get further input from safety experts," the DeepSeek team said in a blog post. "As part of that, we're going to make it available to trusted testers via the Gemini API to get their feedback before making it widely available."

About the Author

John K. Waters is the editor in chief of a number of Converge360.com sites, with a focus on high-end development, AI and future tech. He's been writing about cutting-edge technologies and culture of Silicon Valley for more than two decades, and he's written more than a dozen books. He also co-scripted the documentary film Silicon Valley: A 100 Year Renaissance, which aired on PBS.  He can be reached at [email protected].

Featured

  • robot hand holding stacks of coins

    Designing AI Systems for Financial Aid

    Financial aid offices have been slow to adopt AI, risking technological stagnation at a critical early student touchpoint. Systematic AI integration can improve student experiences and strengthen institutional positioning.

  • digital brain with network connections

    Microsoft Moving to Internally Developed AI Models in Office Apps

    Microsoft is reportedly using its own in-house artificial intelligence models to handle some workloads in Excel and Outlook, offering new evidence that the company is moving its AI strategy beyond model development and into large-scale cost reduction.

  • woman

    AI Giants Back Nonprofit Focused on Workforce Transition

    The AI industry's biggest names are investing in more than just models and infrastructure — they're focusing on workforce readiness. OpenAI, Anthropic, Microsoft and Amazon are backing Raise US, a new nonprofit that aims to raise $1 billion to help American workers prepare for an AI-driven economy.

  • woman surrounded by virtual hologram icons

    Beyond AI Adoption: Designing Learning for an Age of Abundant Intelligence

    Higher education was designed for a world in which access to knowledge, expertise, feedback, mentorship, and authentic learning experiences were inherently scarce. By making many forms of intelligence increasingly abundant, AI is inherently redefining the existing paradigm.