GitHub added Kimi K3 to Copilot on 6 August, lifting a rollout pause it had put in place earlier. That makes five model vendors reachable from one Copilot subscription, ten days after Grok 4.5 made it four.
The model itself will get the attention. The line worth reading twice is the pricing: Kimi K3 is billed under usage-based billing, quoted in tokens, not folded into the premium-request allowance most Copilot users budget against.
What shipped
Kimi K3 is rolling out to Copilot Pro, Pro+, Max, Business and Enterprise — note that plain Pro is included this time, which was not true of Claude Opus 5 in July. It is selectable from the full surface list:
- Visual Studio Code, Visual Studio, JetBrains, Xcode and Eclipse
- The Copilot CLI and the Copilot cloud agent
- github.com, the Copilot app, and GitHub Mobile on iOS and Android
The billing line is the story
GitHub quotes Kimi K3 at $3 per million input tokens, $15 per million output tokens and $0.30 per million cached input tokens, under usage-based billing.
That is a different mental model from the one most teams have built. Premium requests are a unit you can count in your head: a request is a request, and a multiplier tells you what it costs. Tokens are not countable in your head. A long agent run over a large repository can consume a hundred times the tokens of a quick chat and look identical in your activity log.
Two practical consequences. First, cached input at a tenth of the input price makes repeated work over the same repository dramatically cheaper than one-off prompts across many repositories — the shape of your work now changes your bill. Second, if you are on Business or Enterprise, the model is a spend decision as much as a capability decision, and it belongs in whoever owns the budget line, not only in the engineering channel where new models usually get discussed.
GitHub's post does not state a premium-request multiplier for Kimi K3. Treat that as unstated rather than as parity with anything else, and check the billing documentation before you assume.
Business and Enterprise: it is off until you turn it on
Kimi K3 is off by default for Copilot Business and Copilot Enterprise. A plan administrator has to enable the Kimi K3 policy before anyone on the plan sees it in the picker. GitHub explicitly asks admins to review their security and compliance requirements first.
This is now the standard pattern — the same one that applied to Claude Opus 5 and Grok 4.5 — and it is still the single most common cause of "the new model has not appeared for me" tickets. If your developers are asking, the blocker is almost certainly your own admin console rather than GitHub's rollout.
The rollout is also gradual, with GitHub monitoring quality as it goes. An absent entry on day one is not a fault to report.
Should you switch to it?
Not yet, and not on the strength of a changelog entry. A new model in the picker is an option, not an instruction. The teams that get burned are the ones that make a new arrival the default for everyone before anybody has run it on their own codebase.
The cheap way to evaluate it is to take three tasks you have already watched another model do this month — one refactor, one bug with a reproducible test, one piece of unfamiliar code to explain — and run them again. You already know what good looks like on those three, which makes the comparison honest in a way a benchmark score is not.