Cogito

Cogito represents a breakthrough in large language models (LLMs) through its Iterated Distillation and Amplification (IDA) training approach—a self-improving AI methodology that outperforms traditional models. The open-source release includes 3B to 70B parameter models, each surpassing counterparts like LLaMA and DeepSeek in benchmarks, with larger 400B+ models imminent. Unique for its dual-mode operation (direct answers or reflective reasoning), Cogito is optimized for coding, function calling, and agentic tasks.

Developed by Deep Cogito, this project pioneers a path to general superintelligence by overcoming human-supervised limits. Unlike static LLMs, IDA enables iterative capability growth through computational self-reflection. Available via Hugging Face, Ollama, and major APIs, it’s designed for researchers and engineers pushing AI’s frontiers.
Key differentiators:

IDA self-improvement framework

Benchmark-leading open models (3B–70B)

Dual-mode reasoning (direct/reflective)