Moonshot's Kimi K3.1 Pro Ranks Fourth, at a Third Less

Moonshot AI, the Beijing company behind the Kimi chatbot, released Kimi K3.1 Pro on Saturday, 24 October. It is the first model in a new K3.1 series, and Moonshot calls it its most capable model in the 3-trillion-parameter class. It reads up to 1 million tokens at once and, according to Caixin, was released with lower prices and stronger results in coding and agent tasks as Chinese labs try to keep up with a run of upgrades from US rivals.

Moonshot also released K3.1 Pro Highspeed, which it says runs two to three times faster. Both are available on Kimi.com, the Kimi desktop and coding apps, and the Kimi API. The company says the full model weights will be released by 4 November, so anyone will be able to download and run the model on their own hardware.

How it compares

Independent testing puts it close to the top. Artificial Analysis, which runs the same tests across models, scores K3.1 Pro at 55, behind Anthropic's Claude Opus 5.5 at 58, and Claude Sonnet 5.5 and Meta's Muse Wonder Full at 56. Once the weights are out, it will be the highest-scoring open-weights model, cutting the gap to the best closed model from 12 points to 3, according to the firm.

• Terminal-Bench 4.0, which tests coding agents at the command line: 67% in Artificial Analysis's run, up from 13% for Kimi K3. Moonshot's own run gives 65.7%, against 66.4% for Claude Opus 5.5 and 58.2% for GPT-6 Astra.
• Hallucination rate on Artificial Analysis's knowledge test: 44%, down from 53% for K3.
• Cost per task in Artificial Analysis's testing: $1.98, about the same as K3, because the new model writes about 14% more tokens per answer.
• Size: 2.8 trillion parameters in total, with 104 billion active for each token, according to Artificial Analysis.

Moonshot claims bigger gains in areas it measures itself: financial modelling, financial research and building 3D models from text or sketches. Those figures come from the company's internal tests and have not been checked independently.

More from AI and Tech StakApple Finally Puts a Touchscreen on the MacThe new MacBook Pro brings touch, OLED and a Dynamic Island to the Mac from $2,899. What changed, and why Apple gave in.

A model with its cyber skills cut back

Moonshot says it found that the model's skill at finding software vulnerabilities kept rising as training went on, because security content was mixed into its general training data. It says it removed cyber-related data and tasks from the version whose weights it will release, keeping general coding, and that the model's cyber ability remains behind leading closed models. Those are the company's own claims.

A launch under a cloud

The release was not trouble-free. On launch day an X user posted screenshots of what he said was a paid campaign brief asking creators to promote Kimi without labelling posts as paid partnerships, with talking points such as "China just made Opus 5.5 free to use for everyone", Implicator.ai reported. The brief has not been verified, it is not clear who issued it, and Moonshot had not commented as of 25 October.