Run GitHub Copilot on Unlimited Claude

Copilot agent mode rereads files, plans, and edits in a loop, so a single focused session can move a surprising amount of tokens. Put a flat-rate unlimited Claude plan behind it and that volume stops mattering — the price is the same whether the agent makes ten calls or ten thousand.

Bring your own key, without the meter

Copilot's bring-your-own-key flow is what makes this possible. Open the model picker, go to Manage Language Models, Add Models, and choose Custom Endpoint. With an unlimited plan you can wire it once and never think about per-request cost again.

Custom Endpoint replaced the old OpenAI Compatible option. Add one entry per Claude model you want available, knowing every one of them draws from the same flat-rate plan rather than a separate meter.

This changes how you think about which models to register. On metered access people add the cheapest model that will do and leave it there; on a flat-rate plan there is no reason not to register the strongest model alongside it, because using either one costs exactly the same over the period. Registering the full set of Claude models up front costs nothing on this plan and saves you a config trip later.

Point it at the Messages surface

For each model pick the Messages API type and set the URL to the full /v1/messages endpoint. The unlimited gateway serves Claude over the native Anthropic surface, so Messages is the type that matches.

Getting the type right on the first try matters more when you intend to lean on the model. You want the entry solid so that long unlimited sessions never trip on a mismatched wire format halfway through a task.

Spend a minute confirming the entry with a short prompt before you start relying on it. The whole appeal of unlimited is leaving the model running through long work, and a five-second sanity check up front is what lets you trust those long runs not to fail on a misconfigured URL. A solid entry pays for itself across the many long sessions an unlimited plan invites you to run.

Token limits you no longer ration

Each model entry takes an id such as claude-sonnet-4-5 or claude-opus-4-8 and token limits where maxInput plus maxOutput stays within the context window. On metered billing people deliberately shrink these to save money; on unlimited you can set them to the full window.

That difference is the whole point. With the meter gone you stop trimming context to control spend and instead size the request for the best result, because a larger request costs exactly the same as a small one.

Bigger limits translate directly into better answers in Copilot. A generous maxInput lets the agent pull in more of the surrounding code before it responds, and a generous maxOutput lets it return a complete edit instead of a truncated one — both of which you would normally hesitate to allow on a meter. Full-window limits turn out to be one of the most tangible everyday benefits of dropping the meter.

Let agent mode run long

Agent mode needs tool calling, and the gateway forwards Claude's tool calls so it works. The real benefit of unlimited shows up here: an agent that retries, explores, and edits across many files would be unpredictable on per-token pricing.

On a flat-rate plan you simply let it finish. A loop that runs for an hour and a loop that runs for five minutes land on the same bill, so you point the agent at a hard task and walk away.

It also removes the temptation to interrupt the agent early to save money, which often produces worse results than letting it complete. When stopping and continuing both cost nothing, you make that decision purely on whether the work is done, not on what the next batch of tokens will add to the invoice. Letting agent runs complete is, in practice, where most of the value of a flat rate actually lands.

If a model does not show

If a newly added model is missing from the picker, restart VS Code — the list is cached and a restart refreshes it. This is the same quirk everyone hits regardless of plan.

Beyond that, confirm the key works against the gateway and keep the agent-capable model selected for heavy work. Because the plan is unlimited there is no reason to switch down to a weaker model to save tokens.

If the key checks out against the gateway but Copilot still errors, the problem is almost always the entry rather than the plan — an API type that does not match the URL, or a token limit set above the context window. None of these relate to usage, so they are quick to rule out and fix. Because none of these checks touch usage, you can rule them out without ever glancing at a counter.

// VS Code -> model picker -> Manage Language Models -> Add Models
// -> Custom Endpoint  (NOT the deprecated OpenAI Compatible)
// API type:  Messages
// URL:       https://unlimited.aiprimetech.io/v1/messages
// Model id:  claude-sonnet-4-5
// API key:   <your gateway key>
// maxInput + maxOutput must be <= the model context window
// Restart VS Code if the model does not appear in the picker.

Frequently asked questions

Can I run GitHub Copilot on an unlimited Claude plan?
Yes — add a Custom Endpoint in Copilot with API type Messages, URL {SITE}/v1/messages, and a Claude model id. The flat-rate plan covers every request.

Why is unlimited good for Copilot?
Agent mode loops through files and edits, generating high, unpredictable token volume. Unlimited keeps the cost fixed no matter how long a session runs.

My model is not in the picker — why?
Copilot caches the model list. Restart VS Code and it appears. Verify the Messages API type matches the /v1/messages URL too.

Are there token limits on the unlimited plan?
No token limits during the plan, only fair-use rate limits to keep the gateway responsive.

AI Prime Tech is an independent API gateway. It is not affiliated with, endorsed by, or a reseller of Anthropic. Claude and related model names are trademarks of their respective owners.