Open models with a self-improvement thesis
Deep Cogito was founded in San Francisco in 2024 by Drishan Arora, previously a senior software engineer at Google, and Dhruv Malhotra, formerly a product manager at Google DeepMind. The lab's stated goal is general superintelligence, pursued through a distinctive technique called iterated distillation and amplification (IDA) — models search at inference time, then distill those improved reasoning traces back into their own weights, creating a self-improvement loop.
The Cogito model family
Deep Cogito releases its models openly. The Cogito v1 line spanned 3B to 70B parameters, while Cogito v2 scaled to large mixture-of-experts models up to 671B parameters — among the largest open-weight models released by a US startup. The models are hybrid reasoners: they can answer directly or reflect before responding, and are available via the lab's chat interface and open-weight downloads on Hugging Face.
Backing
The company emerged from stealth in 2025 with a $13M seed round led by Benchmark, positioning it as a lean, research-dense counterweight to closed frontier labs.