EXPLORINEER
en

Archive — US — techcrunch

Google rolls out Gemini Deep Think AI, a reasoning model that tests multipleideas in parallel

Google DeepMind has launched Gemini 2.5 Deep Think, its most advanced AI reasoning model capable of exploring multiple ideas simultaneously. Available to Ultrasubscription members, this multi-agent model is designed for creativity and strategic planning, having already scored a gold medal at the Math Olympiad. It outperforms competitors in various benchmarks while utilizing advanced reinforcement learning techniques.

Key points

  • Gemini 2.5 Deep Think is Google's next-gen AI that processes multiple ideas in parallel, improving answer accuracy.
  • The AI model will be accessible to subscribers of Google's $250-per-month Ultrasubscription starting this Friday.
  • Gemini 2.5 Deep Think has demonstrated superior performance in various assessments compared to competition, including scoring 34.8% on the Humanity's Last Exam (HLE).
  • The model uses advanced reinforcement learning techniques and can execute tasks needing longer responses effectively, such as web development.
  • Multi-agent systems use more computational resources but yield better results, which may restrict their use to premium subscription tiers.

Why it matters

  • The launch is significant amid the rising trend of multi-agent AI systems, a concept increasingly explored by tech companies like xAI and OpenAI.
  • Competitors like OpenAI have also highlighted the benefits of multi-agent strategies, indicating a shift in AI model development and use cases.
  • As AI transitions towards more complex reasoning capabilities, it aligns with predictions of adopting more sophisticated AI solutions by 2025.

Google's Gemini 2.5 Deep Think represents a leap forward in AI reasoning capabilities, emphasizing the importance of multi-agent systems in future AI applications and their potential impact on research and development.