Kimi K3.1 Pro: The Open-Weight Giant

Summary:
China’s Moonshot AI has released Kimi K3.1 Pro, a 3-trillion-parameter open-weight model with a 1-million-token context window, built for coding, 3D creation, and financial research. It is live now on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API, with a Highspeed variant, and Moonshot plans to publish the full model weights by November 4. The striking part is what Moonshot left out: it deliberately removed cyber-related data from the training corpus and reinforcement-learning environments, so its internal cyber score fell to 81 percent from 90.5 percent for an unrestricted version — while general coding skills were kept. Moonshot’s stated aim: not to release open weights that would extend the absolute frontier of cyber capability.

Storyline:

  • Hook: China’s Moonshot just released a 3-trillion-parameter AI model anyone will soon be able to run themselves — after deliberately removing its hacking edge.
  • Context: Kimi K3.1 Pro is built for coding, 3D creation, and financial research, is live now on Kimi’s apps and API, and its full weights arrive November 4.
  • Stakes: Open weights put frontier power in anyone’s hands, so Moonshot curbed the cyber-offensive training data first — its internal cyber score fell to 81 percent from 90.5 percent.
  • Payoff: The open frontier is now a safety question, not just a size race: watch what developers build once the weights drop on November 4.

Watch next:
Daytona, the agent-cloud startup, raised a $60 million Series B at a $1 billion valuation led by FirstMark Growth — a bet that AI agents need their own kind of cloud, not the old always-on server model.

Links:

Comments

Leave a Reply