Kimi K3 hits 2.8T params — is memory the new compute?
moonshot ai released kimi k3, a 2.8 trillion parameter model that leverages memory instead of raw compute. this challenges the scaling laws dominating western ai development.
moonshot ai introduced kimi k3, a 2.8 trillion parameter open-weight model launched on july 16. classified in the '3t class', it uses a mixture-of-experts architecture with 2.8 trillion total parameters but roughly 600 billion activated parameters per inference. by focusing on memory optimization over compute, it challenges the prevailing philosophy that bigger compute budgets are the only path to agi.