Moonshot says it removed cyber-related data from training phases for Kimi K3.1 Pro "to reduce potential misuse risk while preserving general usefulness"

The API costs for Kimi's models have fallen by a third, making advanced AI capabilities more affordable.

Moonshot AI has launched Kimi K3.1 Pro, an upgraded version of its Kimi K3 model, achieving a score of 55 on the Artificial Analysis Intelligence Index, an 11-point increase that places it fourth among top proprietary models. The new model demonstrates significant improvements in agentic tasks, coding, and 3D modeling, while its hallucination rate has decreased. Despite maintaining 2.8 trillion parameters, Moonshot has reduced API prices by 33 percent for input, output, and cache operations. Moonshot also stated it has removed cyber-related data from the K3.1 Pro's training to mitigate misuse risks, resulting in a decrease in its performance on specific cybersecurity vulnerability tasks. The company plans to release the model weights by November 2026, which could make it the leading open-weights model.

Moonshot AI's strategic decision to remove cybersecurity-related data from Kimi K3.1 Pro's training corpus, while preserving general utility, highlights a growing industry trend toward safety and responsible AI deployment. This move, however, comes at a quantifiable cost to its performance on specific offensive cybersecurity benchmarks, indicating a trade-off between preventing misuse and maintaining peak capabilities in sensitive domains. The company's stated aim to avoid pushing the frontier of offensive cyber capabilities in open-weights models signals a potential shift in how developers approach the dual-use nature of AI.

The significant price reduction of 33 percent across Kimi K3.1 Pro's API offerings, coupled with its high ranking on the Artificial Analysis Intelligence Index, suggests a competitive strategy to democratize access to advanced AI. This pricing shift, combined with the planned release of model weights by November 2026, could empower a wider range of developers and businesses to leverage sophisticated AI tools. The market should watch how other AI providers respond to this combination of improved performance, reduced costs, and a more cautious approach to potentially hazardous capabilities.