Kimi K3.1 Pro shows open models catching up

26 October 2026 · model

Moonshot AI released Kimi K3.1 Pro on 23 October. It is a mixture-of-experts model with 2.8 trillion total and 104 billion active parameters, a one-million-token context window, and multimodal input. A Highspeed version runs two to three times faster.

Artificial Analysis scores it at 55 on its Intelligence Index, fourth overall. API pricing is $2 per million input tokens and $10 per million output tokens.

The detail worth watching is what Moonshot took out. It says it removed hacking-related material from training before the open release. On its in-house test of exploiting real software flaws, a version trained without those cuts solved 90.5% of tasks and the released model solved 81%. Moonshot plans to publish the full weights by 4 November.

If you build agents, this mix is interesting. It's close to the top models, it's cheap enough for tasks that run for a long time, and the lab chose to give up some ability to make the open release safer.