Renaissance Labs

Atlas / North America

Independent company · San Francisco, United States

Deep Cogito

A San Francisco lab releasing open-weight reasoning models trained by iterated distillation and amplification, with the stated aim of general superintelligence.

language modelspost-trainingreasoning

San Francisco, United States

Founded
2024
Type
Independent company
Based in
San Francisco, United States
Featured projects
4

About

Deep Cogito was founded in San Francisco in 2024 by Drishan Arora, who led language-model work for generative search at Google, and Dhruv Malhotra, a former product lead at Google DeepMind. It describes its goal as building general superintelligence and argues that this requires advances in reasoning and iterative self-improvement.

The lab introduced iterated distillation and amplification (IDA) with the Cogito v1 Preview models in April 2025, followed by Cogito v2 Preview in July 2025. Cogito v2.1, released in November 2025 as a 671 billion parameter open-weight model, was post-trained from an openly licensed DeepSeek base model.

In August 2026 the company announced a $43 million Series A led by TQ Ventures, for total funding above $56 million. Reports say the money will fund hiring, training infrastructure, future open-weight Cogito models and an enterprise business.

Mission and ambitions

Deep Cogito pursues models that improve through repeated cycles of search and distillation. It releases open weights as part of that research program.

Known for

  1. 01

    Iterated distillation and amplification

  2. 02

    Cogito open-weight reasoning models

  3. 03

    Cogito v2.1 release

Milestones

  1. 2024

    Company founded

    Arora and Malhotra started the lab in San Francisco.

  2. 2025

    Cogito v1 Preview

    The lab introduced IDA with its first preview models.

  3. 2025

    Cogito v2 Preview

    A second release moved from inference-time search toward self-improvement.

  4. 2025

    Cogito v2.1

    A 671 billion parameter open-weight model post-trained from an open DeepSeek base, released in November.

  5. 2026

    Series A

    A $43 million round led by TQ Ventures.

Open source

Notable releases

Openness

Open model weights on Hugging Face (several under permissive or llama-style licences).

People

  • Drishan AroraCo-founder and chief executive

    Named in the company's Series A announcement; previously a senior engineer at Google.

  • Dhruv MalhotraCo-founder

    Previously a product lead at Google DeepMind; surname spelling varies across outlets.

Listed from the lab's own pages and public profiles.

Contact and community

Via website.

Sources

Compiled from the lab's public pages, Hugging Face and GitHub. Last checked 2026-11-17. Something wrong or out of date? Tell us on Discord.

Labs nearby