Open-weight reasoning models and custom model training
Raised to date
$40M
of $40M offered across 2 Form D filings
Deep Cogito is a San Francisco AI research company building large language models through post-training and self-improvement methods. It releases the open-weight Cogito models, ranging from about 3 billion to over 600 billion parameters, trained with Iterated Distillation and Amplification, where a model distills its own improved reasoning back into its weights. It also sells an enterprise offering that trains specialized models on a company's proprietary data and outcomes, with Zscaler as a named customer.
Iterated Distillation and Amplification lets models improve from their own reasoning instead of human labels, applied both to open-weight releases and to customer-specific models.
| Filed | Security | Sold of offered | Filing |
|---|---|---|---|
| Apr 24, 2026 | Equity | $40M/ $40M | Form D/A |
| Apr 8, 2026 | Equity | $28.5M/ $40M | Form D |
Large language models
Family of open-weight hybrid reasoning language models from about 3B to over 600B parameters, available through providers such as OpenRouter.
Built for
Developers using open-weight models and enterprises that want domain-specific models
