Moonshot finished the change it announced in August, and it turned out to be mostly a rename. The four paid Kimi tiers are now Plus, Pro, Max and Ultra at the same $19, $39, $99 and $199 a month, the waitlist buttons are gone, and Kimi Code is still inside every one of them. The separate Kimi Code plan that the August banner promised hasn’t appeared: Moonshot’s Kimi Code benefits page still says splitting the two is “coming soon”.
What did change is how much the pricing page tells you. The old plans listed fixed credit counts and a Kimi Code multiplier per tier. The new cards say “More agent credits” on Plus, then “2x”, “5x” and “10x” on Pro, Max and Ultra, and simply “Kimi Code available” on all four. Moonshot doesn’t map the old names to the new ones; the prices match one to one (Moderato was $19, Vivace $199), so that’s how we read it. Annual billing still takes about 20% off, which the page sums up as “Save up to $480” on Ultra.
The part worth understanding before you pick a tier is that two separate limits apply to a Kimi Code session, not one. The first is the membership credit pool, which Moonshot says is shared: “All Kimi membership features share one credit pool.” Run Deep Research in the morning and you have less left for the agent in the afternoon. The second belongs to Kimi Code alone. Its benefits page says Kimi Code credits “refresh automatically every 7 days”, and that “in addition to the weekly credit limit, there is also a 5-hour rolling rate-limit window”. You can hit a wall with credits still in the tank. Neither ceiling is published as a token count.
Moonshot is candid about how soft its numbers are. The help page calls its usage figures “estimates based on typical task token consumption, assuming your monthly credits are used on a single feature.” With a token-hungry model like K3 driving the agent, a heavy refactor consumes a good deal more than a typical task.
Two line items drain credits while you do nothing at all, and both are easy to miss. A deployed Kimi Claw cloud host costs roughly 0.6% of your monthly credits every day, settled at 4 PM, because the sandbox keeps its runtime and data alive whether or not you invoke it. Moonshot’s own fix is to save your files locally and delete the host until you need it again. A website published through Agent costs about 0.08% a day for as long as it stays online, and stops when you unpublish it. Leave a Claw host running for a month and you have spent something like a fifth of your credits on standby.
The escape hatch is the API, billed separately — a membership does not include API usage. Kimi K3 runs $3 per million input tokens, $15 per million output and $0.30 for cache hits, flat at any length up to its 1,048,576-token window. If the work is coding rather than long-context reasoning, though, Moonshot lists a cheaper model that its subscription pages never mention: kimi-k2.7-code, at $0.95 in, $4.00 out and $0.19 cache-hit, described as a coding-focused model for long-context programming tasks. That is under a third of K3’s price. You give up context — 262,144 tokens against K3’s million — and a high-speed variant of the same model costs exactly double. For metered agentic coding, that trade is worth running the numbers on.
The free Adagio tier is not a way in. It costs nothing and gives you one concurrent agent task, two projects and 500MB of storage, but the Kimi Code row on Moonshot’s comparison table is empty for it. Adagio trials the chat and agent product, not the coding tool.
Plus at $19 is the cheapest entry to a frontier-class coding agent anywhere in this tracker, and for bursty work it may be all you need. The case for paying monthly rather than locking in a year is weaker than it was in August, since the plans exist now. It hasn’t gone away: the allowances are relative multiples of a baseline Moonshot doesn’t quantify, and the standalone Kimi Code plan is still “coming soon” with no price.