Moonshot Releases Kimi K3.1 Pro: 3-Trillion-Parameter-Class Open-Weight Model, Deliberately Curbed Cyber Capabilities

What happened

Reuters reported on October 24 that Moonshot introduced Kimi K3.1 Pro on Saturday. The company describes it as its most capable 3-trillion-parameter-class open-weight model: once the weight files are published, outside developers can download them and run or modify the model on their own systems. The company plans to publish the weights by November 4.

The model has a 1-million-token context window, allowing it to process very large amounts of text at once; Moonshot also released a faster version, K3.1 Pro Highspeed, which it said runs two to three times as fast as the standard model.

We aim not to release open weights that would extend the absolute frontier of cyber capability.

— Moonshot AI (via Reuters)

Key facts

1. How it was curbed: cyber-related data was removed from the training corpus and reinforcement-learning environments, while general coding skills were retained (per the company).
2. Quantified cost: on the company's internal cyber benchmark the curbed version scored 81.0% versus 90.5% for the version without risk-reduction measures; the test involved finding and exploiting software vulnerabilities in a controlled environment, and the results have not been independently verified.
3. Weight release: weights are planned to be published by November 4.
4. Regulatory backdrop: China proposed export controls last week that would ban exports of AI-driven malware-generation and mutation technology and restrict AI-assisted vulnerability-discovery tools; the draft is open for public comment, citing risks to critical infrastructure, defense-industrial and core government systems.