Summary

  • Deep Cogito has emerged from stealth with Cogito v1, a range of open-source large language models (LLMs) based on Meta’s Llama 3.2 and fine-tuned with hybrid reasoning capabilities.
  • The company plans to make models that are stronger than those from competitors LLaMA, DeepSeek, Qwen and OpenAI, but will make its models open source.
  • It is seeking to push the boundaries of AI beyond limitations that require human overseers, towards developing superintelligence or AI that is smarter than humans across all domains.
  • The company’s models use a form of self-play, allowing the models to generate improved solutions and then distil their improved reasoning into their parameters, thus creating a feedback look for growth.
  • The company plans to make larger models available in the near future.

By Carl Franzen

Original Article