Overview
- Moonshot announced the Kimi K3.1 Pro launch on Oct. 23 and said it will publish the full model weights on November 4 so developers can run and modify the model locally.
- Kimi K3.1 Pro is presented as a 3‑trillion‑parameter‑class model with a one‑million‑token native context window and benchmarks Moonshot says put it close to leading closed systems for coding, 3D and finance tasks.
- The company introduced a Highspeed variant that it says responds two to three times faster and published API access and per‑token pricing for builders ahead of the weights release.
- To reduce offensive cyber risk Moonshot removed cyber‑related data from training and reinforcement tasks, reporting an 81.0% score on its internal exploit test versus 90.5% for an unfiltered version to show the trade‑off.
- Once the weights are public researchers and security teams will be able to verify Moonshot’s claims, test whether the filtering holds up, and weigh how open distribution changes governance of dual‑use AI tools.