Google DeepMind is launching Gemini 2.5 Deep Think, which the company claims is its most advanced AI reasoning model to date. This system can tackle questions by exploring and evaluating multiple ideas simultaneously before selecting the best answer.
Starting Friday, subscribers to Google’s $250-per-month Ultra subscription will have access to Gemini 2.5 Deep Think through the Gemini app. The model was initially previewed in May at Google I/O 2025 and is the company’s first publicly available multi-agent model. These systems work by deploying multiple AI agents to address a single question in parallel, a process that is computationally intensive but often yields superior results.
Performance and Benchmarks
A variation of Gemini 2.5 Deep Think was used to win a gold medal at this year’s International Math Olympiad (IMO). Google says it is also releasing this specific IMO model to a small group of mathematicians and academics. Unlike consumer-facing AI models that respond in seconds, this specialized version “takes hours to reason.” The company hopes this will advance research and provide valuable feedback for improving the multi-agent system for academic use.
Google notes that the new Gemini 2.5 Deep Think model is a significant improvement over the version announced at I/O. The company also claims to have developed “novel reinforcement learning techniques” to encourage the model to utilize its reasoning pathways more effectively. On Humanity’s Last Exam (HLE), Gemini 2.5 Deep Think scored 34.8% without tools, outperforming xAI’s Grok 4 (25.4%) and OpenAI’s o3 (20.3%). The model also scored 87.6% on LiveCodeBench 6, a competitive coding test, surpassing Grok 4 (79%) and OpenAI’s o3 (72%).
Google says the model automatically works with tools like code execution and Google Search and is capable of generating “much longer responses” than traditional AI models. Internal testing showed the model produced more detailed and aesthetically pleasing web development tasks compared to other AI systems.
Industry Trends and Costs
The multi-agent approach is gaining traction across the AI industry. Elon Musk’s xAI recently released its own multi-agent system, Grok 4 Heavy, which it claims also performs at an industry-leading level. OpenAI researcher Noam Brown has also indicated that the company’s unreleased model used to win a gold medal at the IMO was a multi-agent system. Similarly, Anthropic’s Research agent is powered by a multi-agent system.
However, these systems are more costly to operate than traditional AI models. Consequently, tech companies are likely to reserve them for their most expensive subscription tiers, a strategy both xAI and now Google have adopted.
Google plans to release Gemini 2.5 Deep Think to a select group of testers via the Gemini API in the coming weeks. The company aims to gather insights on how developers and businesses might use the system.
Author’s Opinion
The emergence of multi-agent systems as a new standard for high-performance AI marks a crucial turning point in the technology race. It’s no longer just about who has the biggest or most efficient single model; it’s about who can design a more sophisticated, collaborative architecture. The significant computational costs associated with these systems will likely create a clear divide in the market, with the most powerful and expensive models being reserved for premium users. This trend could accelerate AI development but also widen the gap between those with access to state-of-the-art tools and those without, a dynamic that will be important to watch in the coming years.
Featured image credit: Heute
For more stories like it, click the +Follow button at the top of this page to follow us.