AI labsUnited States
Deep Cogito models
Deep Cogito is an American AI research lab based in San Francisco that builds the open-weight Cogito family of hybrid-reasoning LLMs, including a 671B-parameter MoE flagship. Models are trained via the lab's Iterated Distillation and Amplification (IDA) method, which distills successful reasoning traces back into model weights to give faster, shorter reasoning chains.
- Scored models
- 1
- Best ranked
- Cogito v2.1 671B #278 of 896
- Newest
- Cogito v2.1 671B Nov 19, 2025
- Open weights
- 1 of 1
Deep Cogito models ranked by capability
Capability is Atlas's single score for each model, fitted to every sourced benchmark result (methodology). Rank is among all 896 ranked models from every lab.
| Model | Released | Capability | Rank | Benchmarks | Weights |
|---|---|---|---|---|---|
| Cogito v2.1 671B | Nov 19, 2025 | 137.6 | #278 | 18 | Open |