artificialanalysis.ai Moonshot AI Releases Kimi K3.1 Pro with Reduced Hacking Skills
Article Content
- •Kimi K3.1 Pro scores 65.7% on coding tests, just behind Claude Opus 5.5.
- •Moonshot removed hacking-related training data to limit misuse of the model.
- •The model's hallucination rate dropped from 23.4% to 8.3%.
Moonshot AI has launched Kimi K3.1 Pro, an upgraded version of its AI model, which is set to have its full weights released by November 4, 2026. The model, boasting 2.8 trillion parameters, nearly matches Anthropic's Claude Opus 5.5 in coding tasks, scoring 65.7% on Terminal-Bench 4.0. Notably, Moonshot claims to have removed hacking-related training data to mitigate potential misuse. The model's hallucination rate has significantly decreased from 23.4% to 8.3%. Despite these improvements, K3.1 Pro still trails behind closed models in certain benchmarks. The release comes amid scrutiny from U.S. authorities regarding the use of banned Nvidia chips in its development. Moonshot's proactive measures aim to address safety concerns associated with open-weight models.
Ask AI about this cluster
Answers cite the sources they use
Timeline
More articles in this cluster (4)
Common questions
What improvements does Kimi K3.1 Pro offer?
How does the removal of hacking data affect the model?
When will the full model weights be available?
Continue Reading
Critical BeyondTrust 9.3 Vulnerability Actively Exploited by Threat Actor A critical vulnerability (CVE-2026-102764) in BeyondTrust Remote Support has been discovered and is actively being exploited by the threat group ARMORY CASTLE, reportedly linked to China. The vulnerability, which has a CVSS score of 9.3, allows unauthenticated attackers to upload files and potentially compromise host…