Kimi K3.1 Pro: How Close Is Open AI to the Frontier Now?

Moonshot AI has released Kimi K3.1 Pro, and the gap between open and proprietary language models just got noticeably smaller. Announced on October 23, 2026, K3.1 Pro is the first model in Moonshot's new K3.1 series, and the company describes it as its most capable 3T-class open-weight model to date. It is available today through Kimi.com, Kimi Work, Kimi Code and the Kimi API, and Moonshot says the full model weights will be released by November 4. For researchers and developers who care about running frontier-class AI on their own terms, this is the most significant open-weight release of the month.

• Scale: 2.8 trillion total parameters with 104 billion active, a 1-million-token context window, and text plus image input
• Coding: 65.7% on Terminal-Bench 4.0 by Moonshot's count, versus 66.4% for Claude Opus 5.5 and 58.2% for GPT-6 Astra
• Price: $2 per million input tokens and $10 per million output tokens, down from $3 and $15 for Kimi K3
• Open weights: full release promised by November 4, 2026, with a 2-3x faster Highspeed variant available now

What Is New in Kimi K3.1 Pro?

K3.1 Pro is a post-trained version of Kimi K3, so the underlying architecture and size are unchanged. What changed is the amount of reinforcement learning applied afterward. Moonshot says it rebuilt nearly every layer of its training stack, from the RL framework and long-context inference to sandbox infrastructure and rollout scheduling, so that each additional unit of RL compute delivered stable, predictable gains across hundreds of downstream domains.

The results show up most clearly in agentic work. Moonshot reports 61.2% on its internal Kimi Code Bench 3.0 long-horizon test, up from 55.95% for K3. In one showcase, the model worked through 167 missions to build a human walking simulator on an open anatomical model, diagnosing and correcting 36 errors in shoulder-muscle calculations along the way. K3.1 Pro also works directly with game engines such as Unity, Unreal Engine and Godot.

How Does K3.1 Pro Compare on Benchmarks?

Independent measurement backs up much of the improvement. According to The Decoder, Artificial Analysis gives K3.1 Pro a score of 55 on its Intelligence Index, an 11-point jump over K3 that places it fourth overall, behind only Claude Opus 5.5, Claude Sonnet 5.5 and Muse Wonder Full. Artificial Analysis also measured 1,846 Elo on GDPval-AA, just behind Claude Opus 5.5 at 1,866.

On 3D work, Moonshot reports 92.7% on BenchCAD with tools, close to Claude Opus 5.5 at 96.2%. The company also says its claim-level hallucination rate on a deliberately difficult internal stress test fell from 23.4% for K3 to 8.3%, while noting that the test does not reflect everyday error rates. As with any vendor benchmark, these comparative figures are Moonshot's own unless attributed to Artificial Analysis.

Why the Open-Weight Release Matters

If the weights arrive on schedule, The Decoder notes, K3.1 Pro would become the highest-scoring open-weights model available, shrinking the gap between open and proprietary leaders from 12 points to roughly 3 on the Intelligence Index. That matters for universities, startups and enterprises that need to fine-tune, audit or self-host a large language model rather than rent one.

Moonshot also took a thoughtful step on safety. The company says it found cyber-related content embedded in its general training data, and for the open-weight release it identified and removed that material from both mid-training and the RL stage while keeping the general coding tasks that make the model useful. It is an encouraging example of an open model developer treating capability control as an engineering problem.

What Does Kimi K3.1 Pro Cost?

API pricing drops by 33%: $2 per million input tokens, $10 per million output tokens, and $0.20 for cached input. The Highspeed variant costs $4 input and $20 output for 2-3x faster responses. Moonshot also launched 3D Make on Kimi.com, which lets users preview, edit and export 3D models for printing.

For more on the open model race, explore our AI coverage.

Sources: Moonshot AI — Kimi K3.1 Pro — October 23, 2026; The Decoder — October 23, 2026; The Next Web — October 23, 2026.