Cogito
Family overview — summarizes releases; licenses and availability belong to each release.
Maintained by Deep Cogito16
Cogito is Deep Cogito's series of open-weight language models; the v2 and v2.1 models are hybrid reasoning models that can answer directly or reason first. Cogito v1 Preview (April 2025) covered 3B to 70B models on Llama and Qwen bases; Cogito v2 Preview (July 2025) added 70B dense, 109B MoE, 405B dense, and 671B MoE models on Llama and DeepSeek bases; and Cogito v2.1 (November 2025) is a 671B MoE model with 37B active parameters on a DeepSeek base.13456
- Release notes: Deep Cogito research posts (external site: deepcogito.com)
- Repository: Cogito v2 and v2.1 repository (external site: github.com)
- Release notes: Cogito v2.1 announcement (external site: deepcogito.com)
Releases assessed
| Release | Weights | Tier (USASI rubric v0.1) | License | Released |
|---|---|---|---|---|
| Cogito v2 Preview Llama 405B | Weights: Public | Model-disclosure tier (USASI rubric v0.1): Open-weight | Llama 3.1 Community License Agreement | Jul 31, 2025 |
| Cogito v2.1 671B | Weights: Public | Model-disclosure tier (USASI rubric v0.1): Open-weight | MIT | Nov 19, 2025 |
What it is useful for
Organization context
Provenance and derivatives
All Cogito releases are post-trained from base models released by other developers: Llama and Qwen checkpoints for v1; Llama 3.1 70B and 405B, Llama 4 Scout, and DeepSeek-V3-Base for v2; and DeepSeek-V3-Base for v2.1. Deep Cogito describes its own contribution as in-house post-training using Iterated Distillation and Amplification and process supervision of reasoning chains.357981110
- Derived from: Llama (Meta) — Base models for the Llama-based v1 and v2 releases.
- Derived from: DeepSeek-V3-Base (external site: huggingface.co) — Base model for the 671B v2 and v2.1 releases; not a catalog member.
U.S. eligibility
Eligible · basis: U.S. headquarters
The Cogito models are post-trained and published by Deep Cogito Inc., which is headquartered in San Francisco (see the deep-cogito record). Eligibility rests on Deep Cogito's post-training; the Llama, Qwen, and DeepSeek base models were developed by other organizations. Each release is assessed on its own record.2125
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.