✳ the wire · intel
Kimi K3.1 is coming in August — a Moonshot researcher has confirmed it, and it stays open-weight
An email when every new Moonshot model answers on a public API, about 15 minutes after members, plus every frontier launch · no card
A Moonshot researcher has publicly confirmed that Kimi K3.1 is in progress, with an August target. The claims attached to it: Mythos-level capability retained, faster inference and lower latency, more reliable coding and autonomous agent workflows, fewer unnecessary reasoning steps, and open weights again.
'Fewer unnecessary reasoning steps' is the item to watch. K3's weakness in practice has not been capability, it has been token economics — long deliberation on problems that did not need it. That is a post-training target, and DeepSeek just demonstrated this week what post-training alone can move.
Every wire item is source-verified before it posts and graded: confirmed · strong · reported · rumor. Subscribers see the primary source on each one, the reasoning behind its grade, and the full analysis — plus the launch alerts themselves, which fire the minute a model answers rather than whenever the feed catches up.
