What happened

GitHub Copilot’s model picker, the dropdown where you choose which AI model handles your request, just gained another option. Grok 4.6, xAI’s newest model, is rolling out now across Visual Studio Code, Visual Studio, the Copilot CLI, the Copilot cloud agent, the Copilot mobile app, JetBrains IDEs, Xcode, and Eclipse. GitHub says the rollout is gradual, so it may take a few days to show up in your picker.

It’s available on Copilot Pro, Pro+, Max, Business, and Enterprise plans. If your organization is on Business or Enterprise, there’s an extra step: an administrator has to turn on the Grok 4.6 policy in Copilot settings first. It’s off by default, so nobody on your team sees it in the picker until an admin flips that switch.

GitHub says its own testing found Grok 4.6 performed well on “longer-horizon tasks requiring sustained reasoning and tool use,” the kind of multi-step work where an agent has to hold context across many actions instead of answering one prompt at a time. xAI describes the model as built for that same category of work: navigating a codebase across many steps, not just completing one function.

Why it matters

Copilot now bills usage by the token, the small chunk of text a model reads or writes, instead of counting a flat number per request. GitHub says Grok 4.6 specifically “is billed at provider list pricing under usage-based billing.” Every model on Copilot already draws from your plan’s monthly AI credits, a prepaid balance where one credit equals one cent, based on its own per-token rate. That part isn’t unique to Grok 4.6. What’s worth knowing is that GitHub isn’t negotiating a discounted in-house rate for this one. You’re paying xAI’s own published price for the tokens it processes.

xAI’s own pricing for Grok 4.6 through its API is $2 per million input tokens and $6 per million output tokens, cheaper than several other top-tier models it’s designed to compete with. But GitHub hasn’t published whether the rate it charges through Copilot’s credit system matches that number exactly, so don’t assume it will. Check Copilot’s own pricing table before you lean on the model for heavy agentic work, since a model that runs for many steps in a row can burn through tokens fast regardless of how cheap each one is individually.

If you’re used to a plan’s included usage feeling essentially unlimited for casual work, a long-running agentic session with a new model is exactly the kind of use that can burn through a monthly credit balance faster than you’d expect.

Who should care

This matters if you’re a Copilot subscriber who runs agentic, multi-step tasks, letting the agent work through a larger change on its own rather than accepting single suggestions. That’s the workload Grok 4.6 is aimed at, and it’s also the workload most likely to consume a lot of tokens in one sitting.

It matters less if you mostly use Copilot for inline autocomplete or short one-off questions. Those requests are small regardless of which model answers them, so the pricing distinction here won’t show up in a way you’d notice.

If you administer a Business or Enterprise Copilot account, the decision is upstream of your team: whether to turn the policy on at all, since it stays off until you do.

What builders should do next

Before you switch your default model to Grok 4.6, open GitHub’s models and pricing documentation and find its per-token rate next to whatever model you use today. That comparison, not the model’s benchmark scores, tells you what a long agentic session will actually cost you in credits.

Then run one bounded test. Pick a real multi-step task you’d normally hand to your current model, something like fixing a bug that touches three or four files, and run it once with your usual model and once with Grok 4.6. Compare four things: whether the fix was actually correct, how many follow-up corrections you had to make, how long each run took, and how many AI credits each one consumed. A model that’s cheaper per token isn’t automatically cheaper for the task if it needs more steps, or more corrections, to get there.

End of article