Open model from China: Moonshot releases Kimi K3.1 Pro – with curbed hacking skills
Chinese AI lab Moonshot has released Kimi K3.1 Pro. By its own account it trails Claude Opus 5.5 only narrowly in coding – and cyberattack data was removed from training.
Graphic: TERRA WORLD NEWS (own graphic based on the sources listed)
Chinese AI lab Moonshot on Friday unveiled Kimi K3.1 Pro, an upgraded version of its open model Kimi K3. It is available in the Kimi apps and through the API; the full model weights are due to be released for download by November 4. The context window is one million tokens, and a faster "Highspeed" version is also available.
On the Terminal-Bench 4.0 coding test, K3.1 Pro scores 65.7 percent according to the company – ahead of OpenAI's GPT-6 Astra at 58.2 and just behind Anthropic's Claude Opus 5.5 at 66.4 percent. Moonshot says it cut the share of false claims in an in-house test from 23.4 to 8.3 percent. These figures have not been independently verified.
Before the open release, Moonshot says it removed cyberattack data from training. A test version without this restriction solved 90.5 percent of tasks in an internal hacking test, the released model 81 percent. Open models cannot be recalled once downloaded; safety researchers also warn that safeguards for open models lag behind those of closed ones.