Moonshot says it cut Kimi K3.1 Pro’s cyber score before open-weight release
Moonshot has released Kimi K3.1 Pro in its apps and API, with full model weights due by 4 November. The Chinese AI company says it removed cyber-related training data: on its unpublished in-house exploit test, the released model scored 81%, against 90.5% for an unfiltered version; Reuters reports the figures have not been independently verified.
Bottom line — The 81% result is Moonshot’s own; outsiders will be able to examine the weights after their planned 4 November release.
Go deeper 5
-
Moonshot reports 65.7% on Terminal-Bench 4.0, just behind Claude Opus 5.5 at 66.4%; the company’s figures have not been independently verified, The Next Web notes.
-
The cyber test, xdev-bench-hard, is not public, Implicator.ai reports, leaving outsiders unable to check Moonshot’s comparison yet.
-
Moonshot says K3.1 Pro scored 81% on the exploit test, above K3’s 33.3%; DeepSeek V4.1-Flash scored 85.7% by Moonshot’s count.
-
Implicator.ai reports that Moonshot’s blog gives a hallucination rate of 8.3%, while its API quickstart gives 6.85%; the documents describe different tests.
-
Moonshot says the companion Highspeed model runs two to three times faster than standard K3.1 Pro.