Moonshot launches Kimi K3.1 Pro, with open weights due by November 4

Moonshot announced Kimi K3.1 Pro, which it calls its most capable 3T-class open-weight model: a multimodal mixture-of-experts model with 2.8T total and 104B active parameters and a 1M-token context. Moonshot says K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are live on Kimi.ai, Kimi Work, Kimi Code and the Kimi API (model ids kimi-k3.1-pro and kimi-k3.1-pro-highspeed). Open weights are promised by November 4, 2026.

Moonshot's own claims:

• On Terminal-Bench 4.0 it scores 65.7%, against 58.2% for GPT-6 Astra and 66.4% for Claude Opus 5.5.
• On BenchCAD (CAD code generation from multi-view renderings) it scores 92.7% with tools, against 96.2% for Claude Opus 5.5 and 95.9% for GPT-6 Astra.
• It is state of the art on OfficeQA Pro v2 (64.4).
• On Moonshot's internal benchmarks it gains 7.2 points on financial modeling and 11.6 on financial research over K3, with hallucination in finance work down 65%.
• It builds full projects in Unreal, Unity and Godot and does parametric 3D modeling in Blender and CAD tools.

Moonshot says it removed cyber-related data from training for the open-weight release; on its in-house xdev-bench-hard exploit benchmark the final model solves 81.0% at pass@3, against 90.5% for a checkpoint without those measures, and Moonshot says it remains behind frontier proprietary models on cyber capability.

Moonshot's Kimi Code docs say the K3.1 Pro models support thinking mode only: requests with thinking disabled are rejected with HTTP 400. Highspeed uses about twice the plan quota of the regular model. Prices are not in Moonshot's launch post. Moonshot's API pricing page, which added both models on October 26, lists per million tokens: K3.1 Pro at $2 input, $10 output, $0.20 cache read and $2 cache write; K3.1 Pro Highspeed at $4 input, $20 output, $0.20 cache read and $4 cache write. The launch confirms the rollout that third parties had reported earlier in the day.