Google DeepMind is rolling out Gemini 2.5 Deep Think, its most advanced AI reasoning model yet. The AI can juggle multiple ideas at once, then pick the best answer. It’ll be available Friday for $250/month Ultra subscribers in the Gemini app.
Gemini 2.5 Deep Think is Google’s first public multi-agent model, running AI agents in parallel. This uses more computing power but delivers better results. The model helped Google snag a gold medal at the 2025 International Math Olympiad (IMO).
Google is also releasing the IMO AI model to select mathematicians and academics. This variant spends hours reasoning—unlike typical AI that responds in seconds—to boost research efforts. 
The company says Gemini 2.5 Deep Think is a big upgrade from the version shown at Google I/O 2025. They used “novel reinforcement learning techniques” to improve how the model reasons.
Google stated:
“Deep Think can help people tackle problems that require creativity, strategic planning and making improvements step-by-step.”
Gemini 2.5 crushed benchmarks. It scored 34.8% on Humanity’s Last Exam (HLE), outpacing xAI’s Grok 4 at 25.4% and OpenAI’s o3 at 20.3%. On LiveCodeBench6, a tough coding test, Gemini hit 87.6%, beating Grok 4’s 79% and OpenAI’s 72%.
The model automatically uses tools like code execution and Google Search. It can write much longer, more detailed responses than normal AI.
In tests, Gemini 2.5 crafted cleaner, more appealing web development output than rivals. Google believes it can speed up research and discovery.
Multi-agent AI is the new standard. Elon Musk’s xAI launched Grok 4 Heavy, which also boasts top benchmark scores. OpenAI’s unreleased IMO gold medal AI is multi-agent, too. Anthropic’s research briefs run on a multi-agent system.
These models cost more to run, so companies lock them behind premium subscriptions. Google and xAI are both charging top dollar.
Google plans to open Gemini 2.5 Deep Think to select testers through its API soon. They want to see how enterprises and developers use this multi-agent tech.
The $250 monthly gate keeps Deep Think elite for now—but it’s shaping up as the new AI powerhouse.
