Google has unveiled its most advanced artificial intelligence system yet—Gemini 2.5 Deep Think, a powerful multi-agent reasoning model available exclusively to premium subscribers of the Google AI Ultra plan. Priced at $249.99 per month, the launch marks a bold escalation in the AI arms race, placing Google head-to-head with OpenAI, xAI, and other major players.
This release introduces consumers to Google’s first publicly available AI that uses parallel thinking, a novel method allowing the system to consider multiple solution paths at once before arriving at an answer. Unlike traditional AI models that follow a single line of reasoning, Deep Think evaluates diverse hypotheses simultaneously, enhancing both speed and accuracy on complex tasks.
Outperforming the Competition
Gemini 2.5 Deep Think has already demonstrated superior performance on rigorous testing benchmarks. It scored 34.8% on Humanity’s Last Exam, a wide-ranging AI challenge spanning science, mathematics, and humanities. This surpasses xAI’s Grok 4 (25.4%) and OpenAI’s o3 (20.8%). Additionally, Deep Think outperformed on LiveCodeBench V6, a demanding coding challenge framework, proving its strength in both theoretical and practical tasks.
In a remarkable academic feat, a research variant of Deep Think recently earned a gold medal standard at the 2025 International Mathematical Olympiad (IMO), solving five out of six problems within the standard 4.5-hour time limit—all in natural language. The consumer version, while not as exhaustive in performance, still achieved bronze-level IMO results with significantly faster processing.
A Paradigm Shift in AI Reasoning
At the heart of Deep Think lies Google’s parallel reasoning engine, a breakthrough approach that mimics how multiple minds might tackle a problem collaboratively. This multi-agent system explores several logical pathways simultaneously, drastically improving outcomes for high-difficulty questions and creative problem-solving.
During tests such as the challenging “Catch the Mole” problem from Codeforces, Deep Think maintained active awareness of multiple possible solutions, refining its answer set through strategic queries and optimization algorithms like minimax. This results in more creative, sophisticated, and accurate solutions that would likely elude conventional AI methods.
Access and Features for Power Users
Deep Think is accessible through the Gemini app for Google AI Ultra subscribers, who benefit from the highest usage caps across Google’s AI services, 30TB of cloud storage, and early access to experimental features like next-gen video generation. New subscribers can trial the service at 50% off for the first three months.
In the coming weeks, Google plans to roll out API access to trusted developers and enterprise testers, evaluating how Deep Think might enhance software development, data science, logistics, and beyond.
Looking Ahead
The release of Gemini Deep Think represents a strategic milestone for Google in the global AI race. By combining speed, depth, and creative reasoning, the model may redefine what’s possible in consumer and enterprise AI. As the competition tightens, Deep Think signals that we are entering an era where parallel AI cognition could become the new benchmark for intelligence.
