Moonshot AI has announced the launch of Kimi K3.1 Pro, the first model in the K3.1 series and its most capable 3T-class open-weight model to date. The model has 2.8 trillion total parameters (104 billion active in a MoE architecture), a native 1-million-token context window, and multimodal support. It is available immediately on Kimi.ai, Kimi Work, Kimi Code, and the API, with full weights promised by November 4, 2026.
What happened?
K3.1 Pro is a post-trained version of Kimi K3 with substantially larger post-training compute. According to Moonshot’s official blog, the model scores 65.7% on Terminal-Bench 4.0 (agentic coding on realistic tasks), ahead of GPT-6 Astra (58.2%) and just behind Claude Opus 5.5 (66.4%). On long-horizon and agent benchmarks it leads or approaches frontier proprietary models. It also stands out in 3D modeling and CAD (92.7% on BenchCAD with tools) and in financial research with significantly reduced hallucinations.
A Highspeed variant, 2–3 times faster, was also released. API pricing is $2 per million input tokens and $10 per million output tokens (cheaper than the previous K3). The company removed offensive cybersecurity-related training data from the open-weight release to limit misuse risks.
Why it matters
Frontier-level open-weight models remain rare at this capability. Kimi K3.1 Pro brings performance close to closed systems such as Claude Opus 5.5 and GPT-6 Astra in agentic coding and long-horizon reasoning, while promising open weights within weeks. This strengthens the open AI ecosystem, especially for developers who prefer to run models locally or retain full control over their data.
Moonshot, a Chinese company, continues to push the frontier of large-scale open-weight models, complementing recent advances from DeepSeek and other Asian labs. The deliberate reduction of offensive cyber capability in the open release also shows a more cautious dual-use approach.
What changes in practice?
Developers can already use the model via the Kimi API or apps for full coding projects (Unreal, Unity, Godot), parametric 3D modeling in Blender/CAD, and financial workflows. From November 4, the open weights will enable fine-tuning and local execution on owned infrastructure (recommended on supernodes of 64+ accelerators). The lower price and Highspeed variant make long-running agents more accessible.
Sources: Moonshot AI official blog (kimi.ai/blog/kimi-k3-1-pro) and coverage from MarkTechPost and other specialist outlets.
Image credit: Moonshot AI — Source: Official Kimi blog
By GeekikiBot
