Moonshot AI released its Kimi K3.1 Pro model on Friday, which it claims can keep up with, and in some cases outperform, frontier models like OpenAI's GPT-6 Astra and Anthropic's Claude Opus 5.5.
K3.1 Pro scored 65.7% on the external Terminal-Bench 4.0, which "evaluates coding agents on realistic, resource-constrained terminal tasks using strict task-specific tests." According to its self-published results, the score beats GPT-6 Astra's 58.2% and comes in just behind Claude Opus 5.5's score of 66.4%.
K3.1 Pro is available now through Kimi's website, and Moonshot also released a Highspeed version, which it says runs two to three times faster. The full model weights are expected by Nov. 4, at which point developers will be able to run their own copies.
The release says K3.1 Pro can handle up to 1 million tokens at a time. That's the same as Opus 5.5 and close to the 1.05 million limit on GPT-6 Astra.
Moonshot reported that K3.1 Pro also made improvements in 3D modeling and financial research and made fewer factual errors than its previous Kimi K3 model in an internal stress test.
Moonshot also said it removed cybersecurity training data to keep users from using K3.1 Pro to find and exploit software vulnerabilities ahead of the weights release.