Unlocking Innovation: Google Launches Gemini Deep Think AI for Parallel Idea Testing | TechCrunch

Google DeepMind is set to launch Gemini 2.5 Deep Think, their latest AI reasoning model. It’s designed to think through multiple ideas at once to provide better answers. Subscribers of Google’s $250 monthly Ultra plan can access it starting Friday.

Originally introduced at Google I/O 2025, Gemini 2.5 Deep Think is the company’s first public multi-agent model. This means it uses several AI agents to tackle questions simultaneously, resulting in more thorough answers. In fact, Google used a version of this AI to win a gold medal at the International Math Olympiad.

Alongside the main model, Google is letting select mathematicians and academics test a version used at the Olympiad. This model takes hours to reason compared to other models that work in seconds. Google’s aim is to gather feedback to enhance its use in academic settings.

Google claims that Gemini 2.5 Deep Think is a significant step forward. They’ve also introduced new reinforcement learning techniques to help the AI make better decisions. A representative from Google stated, “Deep Think can support efforts that need creativity and strategic planning.”

The model scored impressively on the Humanity’s Last Exam (HLE), achieving 34.8% without tools. For context, xAI’s Grok 4 scored 25.4%, and OpenAI’s O3 only hit 20.3%. It also outperformed competitors on coding tests, scoring 87.6% on LiveCodeBench6 compared to Grok 4’s 79% and O3’s 72%.

What makes this model stand out is its ability to work with tools like code execution and Google Search. In tests, it produced more detailed web development responses, paving the way for potential breakthroughs in research.

The shift towards multi-agent systems is gaining traction. Elon Musk’s xAI has released a similar system, Grok 4 Heavy, which claims top performance in benchmarks. OpenAI has hinted at using a multi-agent model for its own achievements, echoing the trend among leading AI labs.

However, serving these advanced models comes at a higher cost. Therefore, they may remain exclusive to high-end subscription plans, as evidenced by Google’s approach.

In the coming weeks, Google plans to share Gemini 2.5 Deep Think with select testers through the Gemini API to see how developers might utilize this multi-agent system. This reflects a broader industry move towards more collaborative and powerful AI solutions.

For more insights on AI advancements, check reliable sources like TechCrunch for up-to-date news.

Source link

agents,gemini,Google