Deep Cogito, a San Francisco-based startup founded by ex-Google employees Drishan Arora and Dhruv Malhotra, has emerged from stealth mode with a family of hybrid AI models called Cogito 1. These models can switch between "reasoning" and non-reasoning modes, offering a unique approach to artificial intelligence.
The Cogito 1 models blend reasoning and non-reasoning components to provide efficiency and depth. The reasoning component is similar to models like OpenAI's o1, which excels in complex tasks like mathematics and physics. In contrast, standard AI models are quicker for simpler tasks but may lack in-depth analysis.
This hybrid approach aims to give users the best of both worlds. The models can quickly answer straightforward questions using non-reasoning elements for speed, while dedicating processing time for in-depth analysis of challenging questions. This versatility makes them suitable for a wide range of applications, from quick queries to complex problem-solving.
Cogito 1 models come in various sizes, ranging from 3 billion to 70 billion parameters, with even larger models up to 671 billion parameters planned for release soon. These models are available for download or use via APIs on prominent cloud providers like Fireworks AI and Together AI.
According to internal benchmarking, Cogito 70B outperforms DeepSeek's R1 reasoning model and Meta's Llama 4 Scout model on several mathematics and language evaluations. The company has received funding from firms like South Park Commons, with an ambitious vision to build "general superintelligence."
As the AI landscape continues to evolve, Deep Cogito's innovative approach to hybrid AI models could potentially disrupt the industry and pave the way for more advanced applications. With its unique blend of reasoning and non-reasoning components, Cogito 1 may be poised to make a significant impact in the world of artificial intelligence.
