Moonshot AI has reduced the cyber-offensive capabilities of Kimi K3.1 Pro, its newest model, the Beijing-based startup said on Saturday, Reuters reported. Moonshot describes K3.1 Pro as its most capable 3-trillion-parameter-class open-weight model and plans to publish the weights by November 4.
What Moonshot changed
Moonshot said it removed cyber-related data from the training corpus and reinforcement-learning environments used to develop the model, while retaining general coding skills. “We aim not to release open weights that would extend the absolute frontier of cyber capability,” the company said, according to Reuters.
K3.1 Pro scored 81.0 percent on Moonshot’s internal cyber benchmark, compared with 90.5 percent for a similar version without the risk-reduction measures, the company said. The test involved finding and exploiting software vulnerabilities in a controlled environment. Moonshot said the model still outperformed its predecessor. Reuters said the results have not been independently verified.
The model has a 1-million-token context window. Moonshot also released K3.1 Pro Highspeed, which it said runs at two to three times the speed of the standard model.
Launch quotas
On Monday, Moonshot apologized for rate limiting and unnecessary quota consumption after the launch, Reuters reported. On its Kimi X account, the company said it was offering to reset the monthly quota of all subscribers by 25 percent, which it said equals one week’s usage, Subscribers with usage below 25 percent reset to zero, Reuters reported. Moonshot said it also adjusted parameters to reduce unnecessary subagent calls and output length. The company previously had computing capacity issues after the K3 launch, Reuters said.
Oversight concerns
Open-weight releases let outside developers run and modify a model on their own systems, which makes central oversight more difficult than with a hosted service, Reuters reported. Moonshot argued open-weight models can improve security by helping more organizations inspect and secure their own software. Last week, China proposed export controls that would ban exports of AI-driven malware-generation and mutation technology and restrict AI-assisted vulnerability-discovery tools, according to Reuters.
The launch comes as Moonshot heads toward a planned Hong Kong IPO after a funding round valuing it at about $50 billion, AI Tech Daily reported earlier this month.