THE HOOK
Moonshot just dropped a Chinese open-weight model that trails the best closed one by less than a point, and sanded off its hacking skills first.
THE GIST
Kimi K3.1 Pro is a 2.8T-parameter mixture-of-experts model with 104B active and a 1M-token context. Moonshot reports 65.7% on Terminal-Bench 4.0 against Claude Opus 5.5's 66.4%, and promises full weights by November 4.
THE TAKE
The benchmark gap between open and closed models is now a rounding error. The interesting bit is the self-censorship: Moonshot pulled cyber data from training so the open release doesn't hand everyone a better exploit kit. That's sensible. It's also a promise nobody can check until the weights land, and fine-tuners get their hands on them.
OPINION
Moonshot's own numbers show the cost, with exploit success falling from 90.5% to 81.0% after the cut, per WebProNews. Still well above its predecessor, so "filed down" is doing a lot of work.
- Story type: Breaking news
- Impact: High
- Confidence: High (official announcement)
- Audience: Developers
- What changed: Moonshot released Kimi K3.1 Pro via app and API, with open weights promised by November 4.
- What to watch: Whether the full weights ship by November 4 and match the API model's scores.
- Reading time: 1 minute
