Deep Cogito has emerged from stealth and is introducing a new family of AI models that combine reasoning and non-reasoning capabilities. These hybrid models are called Cogito 1 and will change the face of AI.
Deep Cogito’s solution addresses the long-standing problem in AI: balancing the power of reasoning models with the efficiency of non-reasoning systems. They can methodically work through problems and fact-check themselves along the way. But this advanced reasoning comes at a cost, usually more computing and latency.
Deep Cogito has found a solution, Cogito 1 models, which can switch between reasoning and non-reasoning modes seamlessly. Quick responses to simple queries and deeper, more time-intensive consideration for complex problems. Deep Cogito claims their models outperform the leading open models of similar size, including Meta and DeepSeek. The current Cogito 1 lineup is 3B to 70B parameters, with plans to go up to 671B parameters soon.
About Deep Cogito
What sets Deep Cogito apart is not just the performance of their models but also the speed and efficiency of their development process. A small team built the entire Cogito 1 family in 75 days using existing open-source models from Meta (Llama) and Alibaba (Qwen). Deep Cogito is open-sourcing all Cogito 1 models, with access available via APIs on Fireworks AI and Together AI.
Deep Cogito’s ambitions go far beyond what they have achieved so far. Founded in June 2024 and based in San Francisco, they want to build what they call “general superintelligence” – AI systems that can outperform humans across many tasks and potentially discover new capabilities.
The co-founders, Drishan Arora and Dhruv Malhotra, have experience as industry leaders. Backed by South Park Commons, Deep Cogito is going to make big strides in AI. They acknowledge they are still early in their scaling curve and have only used a tiny fraction of the computational resources used for large language model training. They are actively exploring post-training methods for self-improvement and hint at even more impressive things to come.
Road ahead
Deep Cogito is out of stealth, and the AI community is buzzing, especially around their hybrid AI models. The Cogito 1 family of models is a big deal in AI, combining reasoning and non-reasoning capabilities.
Looking ahead, Deep Cogito will be scaling their models up to 109B to 671B parameters and exploring post-training methods for self-improvement.
As AI continues to move fast, Deep Cogito’s hybrid models are a big step towards more general and powerful AI. Combining symbolic reasoning with neural networks will enable solving complex problem-solving scenarios with unprecedented flexibility and accuracy. This will have a big impact across many industries, from NLP to autonomous systems.
As the AI landscape moves fast, Deep Cogito’s hybrid models are a big deal. By bridging the gap between reasoning and non-reasoning AI systems, they are showing us a future where AI can adapt to the task at hand. This will be super useful across many industries, from scientific research to everyday consumer applications and potentially change how we interact with and benefit from AI.
