Moonshot AI Launches Kimi K3.1 Pro with 1M-Token Context Window

Moonshot AI, a Chinese AI startup based in Beijing, released Kimi K3.1 Pro on October 23, 2026, its most capable 3-trillion-parameter open-weight model to date. The model features a 1-million-token context window and incorporates a mixture-of-experts (MoE) architecture with 104 billion active parameters per forward pass. Moonshot simultaneously released Kimi K3.1 Pro Highspeed, a faster variant delivering 2–3× the speed of the standard model while maintaining comparable intelligence.

The model demonstrates substantial improvements across coding, 3D design, financial research, and multi-step agentic tasks. On Terminal-Bench 4.0, which evaluates coding agents on realistic terminal tasks, K3.1 Pro scores 65.7%, ahead of GPT-6 Astra (58.2%) and narrowly behind Claude Opus 5.5 (66.4%). On BenchCAD, a 3D reconstruction benchmark, it scores 92.7%, approaching Claude Opus 5.5 (96.2%) and GPT-6 Astra (95.9%). The model shows a 65% reduction in hallucination in financial modeling tasks and achieves state-of-the-art performance on OfficeQA Pro v2 (64.4%). On Artificial Analysis Intelligence Index, K3.1 Pro scores 55, placing it well above the median of 27 for comparable models. The model outputs at 73 tokens per second, below the average of 81 for its class.

Moonshot emphasizes cybersecurity safeguards in the model's development. The company explicitly removed cybersecurity-related training data from the pre-training corpus and reinforcement learning environment while retaining general programming capabilities. On Moonshot's internal cybersecurity benchmark, K3.1 Pro scores 81.0%, compared to 90.5% for a comparable version without risk mitigation, indicating a measurable reduction in offensive cyber capability. The company stated: "Our goal is not to release open-source weights that expand the frontier of cyber capabilities." However, the company acknowledged that open-source weight distribution makes centralized regulation more difficult than closed proprietary systems.

The model is available immediately on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API, with full model weights scheduled for release by November 4, 2026. Pricing is set at $2.00 per 1 million input tokens and $10.00 per 1 million output tokens. K3.1 Pro supports text and image inputs and outputs text, with multimodal capabilities. The release positions Moonshot to compete with other leading Chinese AI developers, including DeepSeek and Zhipu, as the company advances toward an anticipated initial public offering in Hong Kong in the first quarter of 2027.