Moonshot Cuts Kimi K3 Context to 256k, Halves the Quota

Moonshot shipped Kimi K3-256k, a cheaper flagship variant capped at a quarter of the context, two weeks after K3 launched, per its docs page.

The smaller model runs at roughly half the quota of full K3 while matching its output inside that 256k window, according to the docs page. That's a real tradeoff, not just a smaller number: Moonshot is selling reduced context as the cost lever instead of hiding it behind vague throttling or pricing tiers that never say what you're giving up.

The docs page positions K3-256k for everyday Q&A, code completion, routine feature work, and single-file edits, the kind of jobs that rarely touch more than a few files at once. It takes image input but not video, and offers low, high, and max reasoning effort settings, with high as the default. The page doesn't say what happens once a task needs more than 256k tokens, only that Moonshot built this version for work that doesn't.

Full K3, a 2.8-trillion-parameter model, launched July 16 with open weights following on July 26. The docs page for the smaller variant picked up 457 points on Hacker News, more attention than most model announcements get for what amounts to a context downgrade.

Pricing a smaller context window as an explicit discount reads more honest than the usual frontier playbook of raising limits and calling it progress. Watch whether other labs start selling narrower windows on purpose instead of only wider ones.