August 2, 2025 - Google DeepMind has launched Gemini 2.5 'Deep Think', a multi-agent AI system capable of parallel reasoning through complex problems. Available exclusively to subscribers of Google's $250/month AI Ultra plan, the model achieved a gold-medal score in this year's International Math Olympiad, demonstrating advanced problem-solving capabilities. The system spawns multiple AI 'agents' to explore solutions simultaneously, though at significant computational cost.
DeepMind claims Deep Think outperforms rivals in coding and knowledge benchmarks, with reinforcement learning optimisations improving its ability to navigate reasoning paths. The company has shared the exact competition model with select academics to advance research in multi-agent AI. As noted in its TechCrunch announcement, Deep Think 'helps people tackle problems that require creativity, strategic planning and step-by-step improvements'.
This release underscores the industry's race to develop reasoning-focused AI, with implications for education, scientific research, and professional problem-solving. The exclusivity to premium subscribers raises questions about equitable access to advanced AI tools, particularly in academic and non-profit sectors.
Our view: Deep Think exemplifies the potential of multi-agent systems to tackle previously intractable challenges. However, its premium pricing model risks exacerbating existing inequalities in AI access. Policymakers must address these disparities to ensure ethical deployment of such powerful technologies.
beFirstComment