Moonshot introduced Kimi K3.1 Pro, a 3T-class open-weight model with a 1M-token context window. It also released Kimi K3.1 Pro Highspeed, offering 2-3× the speed. Both models are available now on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. Full model weights will be released by Nov 4, 2026. K3.1 Pro improves coding, 3D modeling, and financial research capabilities. It scores 65.7% on Terminal-Bench 4.0, ahead of GPT-6 Astra. It scores 61.2% on Kimi Code Bench 3.0 Long-horizon, up from K3’s 55.95%. It scores 92.7% on BenchCAD with tools, near Claude Opus 5.5. It leads OfficeQA Pro v2 at 64.4. Its claim-level hallucination rate on Kimi Hallu Bench Hard is 8.3%. It reaches 81.0% on xdev-bench-hard after cyber-related data removal. K3.1 Pro supports end-to-end 3D creation and parametric modeling. It enables complete financial workflows with improved accuracy.
What changed
Kimi K3 was the prior open-weight model without 1M-token context or these domain-specific gains.
What it unlocks
Running coding, 3D modeling, and financial research tasks with longer context and higher accuracy in one model.
- 1M-token context window
- 65.7% on Terminal-Bench 4.0
- 61.2% on Kimi Code Bench 3.0 Long-horizon
- 92.7% on BenchCAD with tools
What you need to act on it
- access to Kimi.ai, Kimi Work, Kimi Code, or Kimi API
- Kimi Enterprise admins must enable it for their organization
Sources