Moonshot AI unveils Kimi K3.1 Pro with 1M context and open weights Other

Moonshot AI has introduced Kimi K3.1 Pro, an open-weight mixture-of-experts model designed for software engineering, autonomous agents, 3D simulation and financial analytics. The model features a 1 million token context window and is available immediately via the Kimi web interface, Kimi Code and API, with full model weights scheduled for public release by November 4, 2026.

The model uses a mixture-of-experts (MoE) architecture totaling 2.8 trillion parameters with 104 billion active parameters per token. In vendor evaluations, Kimi K3.1 Pro scored 65.7% on Terminal-Bench 4.0, outperforming GPT-6 Astra (58.2%) and trailing Claude Opus 5.5 (66.4%).

What it is

• Long context window. A 1 million token context window allows the model to process large codebases and multi-file project repositories in a single prompt.
• Multimodal tool and engine support. The model works directly with Blender, Unity, Unreal Engine and Godot for 3D modeling, game development and procedural scene generation.
• Long-horizon autonomous problem solving. In simulation experiments, the model ran 167 autonomous iterations on a biomechanical walking simulation, self-corrected 36 calculation errors, and achieved 11 seconds of continuous bipedal walking.
• Reduced hallucination rate. Moonshot AI reports that factual error rates on difficult internal tests decreased from 23.4% to 8.3%.
• Highspeed inference tier. A specialized Highspeed variant operates 2 to 3 times faster than the standard release for interactive coding and fast terminal loops.
• Cybersecurity guardrails. The developers pruned cybersecurity exploit materials from the training dataset to prevent autonomous vulnerability exploitation.

Benchmark results

Sources

• Moonshot AI Kimi K3.1 Pro Announcement