Moonshot AI released Kimi K3.1 Pro on Friday, calling it its most capable 3T-class open-weight model to date. The model has a 1-million-token context window and is available now in the Kimi apps and API, with full weights due by November 4.
Kimi K3.1 Pro is a post-trained version of Kimi K3, built with substantially more post-training compute. Artificial Analysis lists it as a mixture-of-experts model with 2.8 trillion total parameters and 104 billion active, taking text and image input. Moonshot also shipped a Highspeed variant that it says runs two to three times faster.
On the Artificial Analysis Intelligence Index, Kimi K3.1 Pro scores 55, ranking 4th and gaining 11 points over Kimi K3. It trails Claude Opus 5.5 at 58, with Claude Sonnet 5.5 and Muse Wonder Full both at 56. Artificial Analysis said that once the weights are published, Kimi K3.1 Pro will be the highest-scoring open-weight model, narrowing the gap between the leading open and proprietary models from 12 points to 3.
Key facts
- 01Moonshot released Kimi K3.1 Pro on October 23, 2026, with full open weights due by November 4.
- 02The mixture-of-experts model has 2.8T total and 104B active parameters and a 1M-token context window.
- 03Artificial Analysis scores it 55 on its Intelligence Index, 4th overall and 11 points above Kimi K3.
- 04API pricing is $2 input and $10 output per 1M tokens, 33% lower than Kimi K3.
- 05Moonshot says it removed cyber-related training data, cutting an internal exploit score from 90.5% to 81%.
Agentic results drove much of the gain. Artificial Analysis measured 67% on Terminal-Bench 4.0, up from 13% for Kimi K3, and 1,846 Elo on GDPval-AA, 3rd place behind Claude Opus 5.5 at 1,866. Moonshot's own Terminal-Bench 4.0 figures put the model at 65.7%, against 58.2% for GPT-6 Astra and 66.4% for Claude Opus 5.5.
Artificial Analysis lists pricing at $2 per million input tokens, $10 per million output tokens and $0.20 for cache hits, all 33% lower than Kimi K3. It put the cost per Intelligence Index task at $1.98, roughly level with K3, because the new model uses about 14% more output tokens.
Moonshot is pitching the model at full-project coding, parametric 3D modeling and finance. It says Kimi K3.1 Pro can build games in Unity, Unreal Engine and Godot, and a new 3D Make feature is live on kimi.ai and the Kimi app. On Moonshot's internal finance benchmarks, its Business Operation score rose to 37 from 8.9 for K3.
Moonshot also reported a drop in hallucinations. On its Kimi Hallu Bench Hard test, the share of false claims fell to 8.3% with tools, from 23.4% for K3. Artificial Analysis measured a smaller improvement on its AA-Omniscience test, with the hallucination rate falling from 53% to 44%.
The release comes with a deliberate cap on cyber capability. Moonshot said it removed cyber-related data from the mid-training and reinforcement-learning stages for the open-weight version. On its internal exploit benchmark, the released model scored 81%, versus 90.5% for the same checkpoint trained with that data and 33.3% for K3. By Moonshot's count, DeepSeek V4.1-Flash scored 85.7%.
Moonshot said Kimi K3.1 Pro's offensive cyber capability remains below frontier proprietary models and that it does not intend the open weights to extend the frontier of cyber capability. The benchmark figures in Moonshot's post are its own and have not been independently verified.
API customers can call the model as kimi-k3.1-pro or kimi-k3.1-pro-highspeed, with Zero Data Retention available. Kimi Enterprise administrators can enable it for their organizations, where it is off by default at launch.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
