← Back to Blog Explainer

What Are GitHub Copilot AI Credits?

Published: July 2026

GitHub Copilot AI Credits are a usage-based billing unit that replaced Copilot's old "premium requests" model on June 1, 2026. Every plan includes a monthly credit allotment, and usage is metered on token consumption — input, output, and cached tokens — at each model's listed API rates. Paid plans can buy more credits when the pool runs out.

If your Copilot bill suddenly looks different this summer, you are not imagining it. On June 1, 2026, GitHub retired premium-request counting and switched to metered AI Credits. Here is exactly what changed, what still costs nothing, and how to keep your usage inside the pool.

What changed on June 1, 2026

Before June 2026, GitHub Copilot measured paid usage in "premium requests" — a flat unit where one chat message or one agent action counted as roughly one request, regardless of how large it was. That model was simple, but it hid a lot of variance: a one-line question and a 50-file agent refactor both counted the same, even though the refactor cost GitHub far more to run.

The new AI Credits system fixes that by billing on what actually drives cost: tokens. Every credit-consuming request is now metered on its real token consumption — input tokens, output tokens, and cached tokens — priced at each model's listed API rate. A frontier model reasoning over a large context burns credits quickly; a lightweight model answering a short question barely moves the needle.

This is the same token-based logic that governs raw API billing. If you have compared the pricing of AI coding tools or looked at what agent sessions actually cost, the shift will feel familiar: Copilot is now closer to a metered API than a flat subscription.

What is still free

The most important detail for most developers: your day-to-day autocomplete did not get more expensive. Two features remain free and draw zero AI Credits on every plan, including the Free tier:

If you mostly ride autocomplete and only occasionally open chat, the June change is essentially invisible to you. Your credits sit largely untouched month to month.

What draws down your credit pool

Three features pull from your monthly AI Credits allotment:

Because billing is token-based, the same feature can cost wildly different amounts depending on the model and context size. An agent task on a frontier model with a long file tree in context can consume many times the credits of the same task on an efficient model with a tight prompt. Model choice and context discipline are now direct cost levers — exactly the same dynamics you see in general GitHub Copilot pricing.

Plan allotments at a glance

Every plan ships with a monthly pool of GitHub AI Credits. Paid plans can purchase additional credits once that pool is spent. Here is where each plan lands as of July 2026:

PlanPriceMonthly AI Credits Included
Free$0Limited monthly allotment
Pro$10/mo$10 in credits
Pro+$39/mo$39 in credits
Max$100/mo$100 in credits
Business$19/user/mo$19 in credits
Enterprise$39/user/mo$39 in credits

Note the pattern on paid individual and organization plans: your monthly credit allowance is denominated to match your subscription price. A Pro seat at $10/mo includes $10 of credits; a Business seat at $19/user includes $19 of credits. Beyond that, usage bills as overage at the underlying token rates.

One promotion worth knowing: Business and Enterprise plans received 2x promotional credits through August 2026, effectively doubling the included pool for organizations during the transition. If you are budgeting an org rollout, plan for that promo to end and true usage to surface in the fall.

What it means for you in practice

The practical takeaway splits cleanly by how you use Copilot.

Light users are unaffected

If your Copilot usage is 90% inline completions, nothing meaningful changed. Completions and Next Edit suggestions are free, so your credit pool barely depletes. You will likely never see an overage line.

Heavy agent and chat users feel it

If you live in agent mode on frontier models, the picture is different. Token-metered billing means a handful of large agent runs can drain a monthly pool, after which every additional request bills as overage at API token rates. The developers most affected are precisely those getting the most value — which is exactly why managing token consumption now matters.

Where your credits go in a typical agent request

System prompt & tool definitionssent every call
Your prompt textyou control this
File contents & context injected into the runoften the largest slice
Model output tokenspriced highest per token
Every token here is metered against your pool 

How to manage your AI Credits

You do not need to abandon agent mode to stay inside your allotment. A few habits keep token consumption — and therefore credit burn — under control:

The last two points are where prompt discipline pays off. Since AI Credits are metered on tokens, anything that reduces the tokens you send reduces what each credit-drawing request costs. Terse compresses your prompts on-device before they reach the model and tracks per-request cost, so the same agent task and the same chat message spend fewer tokens — and fewer credits — for the same result. You can estimate the effect yourself with the token calculator.

If you specifically want to trim Copilot token usage, our guide for GitHub Copilot users walks through the setup end to end.

Make every credit go further

Terse compresses prompts in real time and tracks per-request token cost across Copilot chat, agent mode, and every other AI tool you use. On-device, zero latency, no API calls — so your AI Credits stretch across more real work.

Download Terse

The bottom line

GitHub Copilot AI Credits turn Copilot into a metered, token-based product. Completions stay free; agent mode, chat, and code review draw from a monthly pool sized to your plan. Light users will not notice. Heavy agent users on frontier models will — and their best defense is the same one that governs any token-billed system: choose models deliberately, keep context lean, compress prompts, and watch the meter.

Further Reading

Related articles

How Many Tokens Do AI Coding Agents Use? Why Is Cursor So Expensive? Why Is Claude Code So Expensive?