2026-10-24 / Signal #2
Moonshot dropped a 3-trillion-parameter open model three points from the frontier — then deleted the hacking skills
“The open model that almost caught Claude — except they taught it not to hack first.”
The Story
Kimi K3.1 Pro (2.8T total / 104B active MoE, 1M context) scores 55 on the independent Intelligence Index, three points behind Claude Opus 5.5 and far above prior open models. Full weights drop November 4. Moonshot deliberately stripped cyber-attack training data from the open release, costing it points on its own exploit test.
Why It Matters
Open vs. closed economics plus a strange cultural choice: a Chinese lab publishing near-frontier weights while surgically removing the “goblin” capability. Builder utility (what you can actually run) collides with identity/law (who gets the dangerous skills).
Evidence
Moonshot/Kimi tech blog (Oct 23); Artificial Analysis Index; The Decoder; HN (949 pts).
Sources
Daily scan: 2026-10-24