The new SWE-2 model was introduced by Cognition, built on the Kimi K3 base with 2.8 trillion total parameters and 104 billion active per token through a mixture-of-experts architecture.
register or log in to get more services such as weather report, extended news articles and web search