New AI Company Deep Cogito Releases First Hybrid AI Models

The new American AI company Deep Cogito released its first open large language models (LLMs) called Cogito v1 on Tuesday. The startup claims its open-source and hybrid models outperform similar open AI models, including DeepSeek, Qwen, and Meta’s LLaMA.

Deep Cogito, headquartered in San Francisco, introduced Cogito v1 through an official announcement on its website, including details of its LLMs, its performance, and upcoming releases.

“We are releasing the strongest LLMs of sizes 3B, 8B, 14B, 32B, and 70B under open license,” states the document. “Each model outperforms the best available open models of the same size, including counterparts from LLaMA, DeepSeek, and Qwen, across most standard benchmarks. In particular, the 70B model also outperforms the newly released Llama 4 109B MoE model.”

The new model has been trained with the Iterated Distillation and Amplification (IDA), a framework that trains AI models to build aligned systems for general superintelligence, and includes “reasoning” features. All models can function in reasoning and standard mode—a hybrid modality—and have been optimized for function calling, coding, and agentic use.

According to the startup’s research and results, Cogito v1 outperformed most equivalent models in direct and reasoning modes, considering popular benchmarks. The company expects to launch larger models in the next few months.

“Our next release will feature updated checkpoints for each model size (3B to 70B), with extended training periods, as well as larger models in the coming weeks and months,” wrote the company. “All models will be open source.”

Cogito v1 can be downloaded on Ollama or Hugging Face, or accessed through APIs on Together AI or Fireworks AI.

Cogito v1 Preview is now live on Together AI ✨

These open models from Deep Cogito push the boundaries of reasoning and alignment, with options up to 70B parameters.

Available now via Together Dedicated Endpoints. pic.twitter.com/hc0gYXU7x1

— Together AI (@togethercompute) April 8, 2025

According to TechCrunch, Deep Cogito was founded in June 2024 by Drishan Arora and Dhruv Malhotra—both former Google employees. The company is backed by South Park Commons, and its main goal is to build a general superintelligence that can outperform humans.

The new AI model arrives to keep up with the intensity and speed of the AI market and to join the open-source trend. The Chinese search engine giant Baidu announced a new upcoming open-source AI model in February, and DeepSeek shared more details and a transparency initiative for its code just a few weeks ago.