Deep Cognito, the latest AI research startup based in San Francisco, has emerged from stealth along with Cogito 1, a comprehensive suite of open-source AI models.
It can be toggled between “reasoning” and non-reasoning modes.
All of Deep Cogito’s models, called Cogito 1, are hybrid models. Cogito claims that they outperform the best open models of the same size, including models from Meta and Chinese AI startup DeepSeek.
All the Deep Cognito’s models, also known as Cognito 1, are hybrid models, which can instantly answer simple questions while spending extra time considering more challenging queries.
The company stated, “Each model can answer directly or self-reflect before answering (like reasoning models), All were developed by a small team in approximately 75 days.”
The Cognito 1 models range from three billion parameters to 70 billion parameters. In addition, models ranging up to 671 billion parameters will join them in the near future.
Deep Cogito was developed on top of Meta’s open Llama and Alibaba’s Qwen model to create its own.
Notably, the company has implemented novel training approaches for improved base models’ performance.
The results of Cognito’s internal benchmarking, the largest Cogito 1 model, Cogito 70B, with reasoning, exceeds the Chinese AI model, DeepSeek’s R1 reasoning model, on some mathematics and language evaluation.
Each Cognito 1 model can be installed through APIs on cloud providers Fireworks AI and Together AI.
