On December 2, 2025, GitHub started removing legacy $0 budgets for GitHub Copilot premium requests on enterprise and team accounts created before August 22, 2025. That sounds minor until a few engineers hit "run again" on a reasoning model and your allowance silently shifts into paid usage. The good news: you’re in control. With the right policy toggles, budgets, and reporting, you can keep the lights on for power users without waking your finance team.
What exactly changed on December 2, 2025?
Beginning December 2, GitHub is removing the automatically created account-level $0 budget that many enterprise and organization accounts relied on to block premium request overages. After removal, overages are governed by your premium request paid usage policy (Enabled or Disabled) plus whatever budgets you set. Pro and Pro+ individual plans are not affected by this budget removal, but organizations and enterprises are.
Three immediate implications:
- If your policy is Enabled and you have no blocking budget, users can exceed their monthly allowance and incur per-request charges.
- If your policy is Disabled, usage stops at the allowance—no surprise spend, but power users will hit a wall.
- Budgets still work. You can set a bundled budget (all Copilot AI tools together) or SKU-specific budgets (per tool) to cap overages.
How GitHub Copilot premium requests work
Every paid Copilot plan includes a monthly allowance of premium requests, which are consumed by advanced models and certain features like agentic workflows. Allowances reset on the first calendar day of each month, not on your subscription renewal date. If you enable overages, each premium request beyond the allowance is billed at $0.04, sometimes with a model multiplier depending on the model or feature you choose.
Current plan allowances:
- Copilot Pro: 300 premium requests per month.
- Copilot Pro+: 1,500 premium requests per month.
- Copilot Business: 300 premium requests per user per month.
- Copilot Enterprise: 1,000 premium requests per user per month.
What’s unlimited: code completions and chat/agent mode with included models remain unlimited for paid plans, though rate limits still apply. Premium requests kick in when you select higher-end or specialized models, agent features, or certain multi-file or code-review flows. Usage varies by model and feature, so expect heavier hits from more capable reasoning models.
Who’s affected—and who isn’t?
Affected: Enterprises and organizations with an account-level $0 premium request budget created before August 22, 2025. These legacy $0 budgets are being removed, shifting control to the policy switch and any budgets you configure.
Not affected by the removal: Individuals on Pro or Pro+. They still have per-user monthly allowances and can optionally buy overages. Their default overage budget remains $0 unless they change it in personal billing settings.
“Will completions or chat stop working if we hit the limit?”
No—paid plans keep unlimited completions and unlimited chat/agent mode with included models. What stops (or turns paid) is premium-model or premium-feature usage once the allowance is exhausted, depending on your policy and budgets. If you’ve set the policy to Disabled, those premium actions are blocked. If Enabled with budget headroom, they’re billed per request.
Show me the numbers: what can overages cost?
Here’s a conservative way to reason about cost without overengineering a spreadsheet. Start with a typical team of 50 engineers on Copilot Business. Each has 300 included premium requests a month, so your included pool is 15,000. Assume your top 20%—10 people—regularly exceed that. If each does 400 extra premium requests in a busy month, that’s 4,000 paid requests. At $0.04 each, that’s $160. For most teams, that’s well worth the lift in velocity.
Now scale to a 500‑dev enterprise with heavier usage. Suppose 20% do an extra 1,000 premium requests in a crunch month. That’s 100,000 paid requests, or $4,000. Still manageable—until a few users hammer reasoning models that carry multipliers. Your goal isn’t to forbid those users; it’s to set a practical ceiling, monitor spike patterns, and steer heavy users to the right plan tier.
60‑minute admin runbook for this week
If you have limited time, run this play in the next hour and you’ll be covered.
- Confirm the policy switch: In enterprise or org Copilot settings, locate “Premium request paid usage policy.” Decide: Enabled (allow overages) or Disabled (block overages). If you’re unsure, set Disabled first, then proceed to budget setup so you can enable with guardrails.
- Set a bundled monthly budget: Choose a number that reflects a tolerable overage for December—e.g., 10–20% of last month’s observed premium usage. Enable “stop when budget is reached.” This creates a soft landing: users get some extra headroom but can’t blow past a cap.
- Turn on SKU budgets for hot tools: If your org uses the coding agent or code review heavily, create smaller per‑SKU budgets so one tool can’t drain all headroom.
- Pull the usage report: Download last month’s premium request report. Identify the top 10 users by premium usage and the top 3 orgs by total consumption. Flag anyone who exhausted their allowance in week one.
- Adjust the outliers: For power users consistently maxing out Business, consider moving them to Copilot Enterprise. Its 1,000 monthly premium requests often pencils out versus paying overages for Business users who exceed ~800 premium requests regularly.
- Set alerts: Configure alerts at 75%, 90%, and 100% of your budget. Ask team leads to nominate a point person to receive the 90% alert and triage with you.
- Communicate a 2‑line policy: Post in your engineering channel: “Use any model you need. If you hit a block, ping #copilot-admin for a temporary budget increase or a one‑click plan upgrade.” Keep it developer‑friendly.
Should we disable overages entirely?
Use this quick framework.
- Disable overages if your finance policy requires hard monthly caps, you haven’t reviewed usage data, or the team is new to premium models. Pair Disabled with a modest bundled budget set to $0 for the first week, then revisit after you have data.
- Enable overages with budgets if you have a baseline on usage, want to avoid blocking key workflows, and can tolerate a predictable ceiling. This is the most common enterprise setup.
- Enable overages without budgets only if your teams are seasoned, you’ve instrumented alerts, and the spend is a rounding error compared to the value delivered. Most orgs shouldn’t start here.
What counts as a premium request?
Premium requests are consumed by higher-end models and certain agent features. The exact usage depends on the model and feature. Reasoning models and multi‑step agents generally use more per interaction than lightweight models. If you’re rolling out code review and the coding agent, expect premium request usage to cluster around those workflows. That’s why SKU budgets are useful: they let you throttle or prioritize by tool.
People also ask: common admin questions
Do chat messages consume premium requests?
Chats with included models don’t consume premium requests for paid plans, but switching to premium models or agent work can. The UI will usually indicate when a premium model or feature is in use.
When do allowances reset?
On the first day of every calendar month. If you made changes in late November, your counters reset on December 1, 2025.
How can developers see their usage?
Usage shows up in VS Code, Visual Studio, JetBrains, and on github.com. Ask developers to check the Copilot status icon or their account usage page; admins can download enterprise or org‑level reports.
What if we have multiple billing entities?
Users with licenses from multiple orgs or enterprises must pick which entity is charged for premium requests. Until they choose, premium requests may be rejected. It’s worth sending a short PSA to those users.
Model strategy: put the right horsepower where it pays back
I’ve seen teams get the most ROI by pairing lightweight models for quick prompts with premium reasoning models for sticky refactors, gnarly bug hunts, or cross‑repo changes. Don’t try to police model choice line by line; instead, offer guidance on when to switch up. A short internal guide—“When to use reasoning models vs. fast models”—beats a thousand approvals. Encourage devs to land bigger changes in branches and let the coding agent handle the mechanical work while humans review.
Chargeback and forecasting with SKU budgets
Dedicated SKUs for Copilot AI tools make budget hygiene easier. Start by tagging each SKU budget to a cost center (core platform, product A/B, shared services). In monthly reviews, identify which products consume the most premium requests and correlate with delivery metrics: pull request throughput, MTTR, and escaped defect rates. You’ll quickly see whether premium model usage is funding velocity or just fueling chatty experiments.
Governance without friction
Here’s the thing: rigid controls will push developers to alternative tools. A better pattern is “default open, budgeted caps.” You allow premium features, cap the blast radius with budgets, and move habitual heavy users to the right plan. Publish a one‑page policy, include the appeal path, and rotate a small admin crew who can grant temporary increases within 24 hours. That keeps trust high and blockers low.
Checklist: your repeatable monthly cadence
- Day 1–2: Download last month’s usage by org and by user. Flag top 10% by premium consumption.
- Day 3: Move qualified heavy users (e.g., >800 premium requests on Business) to Enterprise.
- Day 4: Recalibrate bundled and SKU budgets. Keep “stop at budget” on.
- Day 5: Send a note with model‑use tips and the #copilot-admin contact.
- Mid‑month: Review spikes and temporarily lift a SKU budget if a release crunch needs it.
What to do next (this week)
- Decide your policy: Enabled with budgets, or Disabled for a hard cap. Make the call today.
- Set bundled and SKU budgets, and turn on alerts at 75/90/100%.
- Pull usage reports, tag heavy users, and upgrade the few who consistently exceed Business allowances.
- Post your two‑line policy and office‑hours channel. Keep it friendly and fast.
- Schedule a 20‑minute review after the first week of December to check spend and unblock teams.
Related deep dives and help
If you want more step‑by‑step guidance, we’ve been tracking these changes closely. Our breakdown of the rule changes is in this Dec 2 rulebook. For a shorter, here’s‑what‑to‑do‑today read, see the Dec 2 switch checklist. And if your org needs a fast policy setup or usage review, our team can help—start at services or drop a note via contacts.
Zooming out: why this is a net positive
Budgets and policies aren’t a tax on innovation—they’re what let you scale it safely. GitHub’s move away from static $0 budgets reduces confusing edge cases as new AI tools arrive and gives you a cleaner lever: enable or disable paid usage, then cap it. With a light monthly rhythm and SKU budgets, you’ll keep the fast lanes open while keeping finance looped in. That’s how you make AI assistance boring in the best possible way—reliable, predictable, measurable.
FAQ the team will ask you
Can we see which model burned the budget?
Yes. Usage reports and analytics break down premium requests, including by dedicated SKUs. That’s enough to spot whether code review or a coding agent caused a spike.
Can we let a single org within the enterprise have a bigger budget?
Yes. Budgets can be set at enterprise, organization, and sometimes at cost center or member level. Teams running hot can get bigger caps without exposing the whole enterprise.
If we disable overages and something critical is blocked, how fast can we fix it?
Immediately. An admin can bump a budget or flip the policy for a short window while work completes. Pair that with an alert and you’ll avoid unplanned downtime.
Final word
December 2 was the nudge to put proper controls around premium models. Take an hour to set policy, budgets, and alerts; move a handful of heavy users to the right tier; and write a two‑line internal policy. After that, let teams build. If you want a second set of eyes, we’ve helped clients implement this in a day—our blog has more playbooks, and we’re a message away.
