The news: AI startup Deep Cogito has exited stealth with Cogito 1, a family of open-source hybrid AI models trained to toggle between fast-response and step-by-step reasoning modes.
Unlike OpenAI’s slower-but-more-precise reasoning models (like o1) or Meta’s Llama (fast but less nuanced), Deep Cogito says its models outperform both in speed and accuracy, per TechCrunch.
Built atop Meta’s Llama and Alibaba’s Qwen, Cogito 1 scored higher than Meta’s Llama 4 and DeepSeek’s R1.