Chinese startup Moonshot AI has released Kimi K3.1 Pro, a new model in the 3 trillion parameter class featuring a one million token context window. The model is designed for coding, 3D development, and financial research, with a high-speed variant available that offers response times two to three times faster than the standard version.
In performance testing, Kimi K3.1 Pro scored 65.7% on Terminal Bench 4.0, nearing Claude Opus 5.5 at 66.4% and exceeding GPT 6 Astra at 58.2%. The model also achieved 92.7% on BenchCAD for 3D modeling and 61.2% on the internal Kimi Code Bench 3.0, showing progress from the 55.95% score of its predecessor, Kimi K3.
The model demonstrates improved accuracy, with internal hallucination rates dropping from 23.4% to 8.3%. Moonshot also adjusted the training process to mitigate cybersecurity risks, resulting in an 81% score on internal vulnerability benchmarks compared to 90.5% for an unrestricted version.
Kimi K3.1 Pro is currently accessible via Kimi.ai, Kimi Work, Kimi Code, and the company's API. Moonshot plans to release the full model weights publicly on November 4, enabling developers to deploy the technology on their own infrastructure.
