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.
log in to read full article
the world wide web for retro machines
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.
log in to read full article