GitHub Copilot Premium Requests: The Dec 2 Switch

Here’s the thing: on December 2, 2025, GitHub flips a billing switch for many organizations using Copilot. The auto‑created $0 budgets that used to block premium request overages are being removed for older Enterprise and Team accounts. After the removal, whether you pay beyond your included allowance depends entirely on your “premium request paid usage” policy. If you’ve never looked at that setting, today’s the day. (github.blog)

What are “GitHub Copilot premium requests,” really?

Copilot splits activity into two buckets: unlimited usage of the included models (useful for completions and chat) and “premium requests” that meter advanced models and features like the coding agent, Spark, and code review. Each premium request counts against a monthly allowance tied to your plan; if you allow overages, extra usage bills at $0.04 per request, with multipliers for certain models. Think of multipliers as a unit converter: a 10× model turns one prompt into 10 premium requests. (docs.github.com)

Allowances are not hand‑wavy. GitHub’s plans page calls them out explicitly: 50 premium requests for Copilot Free, 300 for Pro, 1,500 for Pro+, 300 per user for Business, and 1,000 per user for Enterprise. Counters reset at 00:00:00 UTC on the first of the month. (docs.github.com)

What exactly changes on December 2?

GitHub is removing account‑level $0 Copilot premium request budgets for Enterprise and Team accounts that were created before August 22, 2025. Once removed, your policy determines whether premium request usage can exceed the plan’s allowance—and therefore whether charges accrue. GitHub flagged this change in their changelog and community announcements, and they’ve also introduced dedicated SKUs per AI tool (for example, coding agent and Spark) that improve tracking by tool. (github.blog)

Two immediate consequences: if your policy is enabled (allow overages), you can incur costs beyond your allowance; if disabled, premium features shut off once a user hits their allowance—while basic completions and chat on the included models keep working. (github.blog)

“Will Copilot break if we disable premium request overages?”

No. Paid plans keep unlimited code completions and chat with included models (subject to rate limits). Premium features—the ones that consume the metered units—are what get blocked at the allowance boundary if overages are disabled. If you’ve seen developers relying on Claude‑class models, the coding agent, or Spark, they’ll be the ones who notice. (github.blog)

Premium request allowances and pricing at a glance

Let’s anchor on the numbers from GitHub’s public materials:

• Allowances per month: Pro 300, Pro+ 1,500, Business 300 per user, Enterprise 1,000 per user; Free includes 50. (docs.github.com)
• Price beyond allowance: $0.04 per premium request, multiplied by the model’s factor (e.g., 10× for certain “opus‑class” models; 1× for Gemini 2.5 Pro; 0× for included models like GPT‑4.1/4o on paid plans). (docs.github.com)
• Reset schedule: monthly on the 1st at 00:00:00 UTC. (docs.github.com)

Practical read: most developers won’t hit limits if they stick to included models, but power users working with agents or premium models will chew through allowances quickly—especially when multipliers enter the chat. (docs.github.com)

Where to find and fix the setting (5‑minute admin walkthrough)

Here’s the fastest way to avoid surprise charges or, if you choose, enable them with intention:

1. In GitHub, open your enterprise or organization settings and navigate to Copilot → Spending. Look for “Premium request paid usage” and the budgets section. (docs.github.com)
2. Decide your stance: set the paid usage policy to Disabled (block overages) or Enabled (allow overages). This policy governs behavior after included allowances are exhausted. (docs.github.com)
3. If you want a hard cap, create or adjust a Bundled premium requests budget. When the budget is reached, premium‑metered tools stop for everyone in that billing entity. (docs.github.com)
4. Download your usage report to see who’s actually consuming premium requests. You’ll likely find a long tail plus a handful of heavy users. (docs.github.com)
5. Optionally upgrade the heaviest users to Copilot Enterprise to raise allowances if that’s more cost‑effective for them. (docs.github.com)

Cost modeling you can trust (with multipliers)

Let’s get practical. Suppose you’ve got 100 developers on Copilot Business and you enable overages. Ten of them are power users who each send 500 premium prompts to a 1× model in December, after burning through their 300 included requests. That’s (500 − 300) × 10 = 2,000 overage requests, or $80. Now swap in a 10× model for 20 of those prompts: each one costs 10 premium units, so those 20 prompts consume 200 premium requests per user. Your overage jumps accordingly. The math’s simple, the surprises aren’t. (docs.github.com)

Here’s the mindset: premium requests are a shared utility. Set a policy, cap it with a budget when you need hard edges, and funnel your known heavy users to the plan that matches their workload. GitHub’s docs even hint that some Business users making ~800 premium requests a month could be cheaper on Enterprise. (docs.github.com)

Why is GitHub removing $0 budgets now?

GitHub is introducing dedicated SKUs per AI tool, starting with the coding agent and Spark. Keeping per‑account $0 budgets in place would force admins to clone and manage multiple budgets as new tools land. Moving to a policy‑driven model simplifies that administrative mess and gives finer control per tool. If you’ve felt Copilot’s feature velocity over the last six months, the rationale tracks. (github.blog)

GitHub Copilot premium requests: governance that doesn’t slow teams

We use a three‑layer guardrail model with clients. It takes 30 minutes to implement and saves you hours of cleanup later.

1. Policy default: Set premium request paid usage to Disabled at the enterprise level. This keeps the lights on (completions + included chat) while preventing unbounded spend. Create a named exception org for teams who truly need overages. (docs.github.com)
2. Budget rails: For any entity where you enable overages, set a Bundled premium requests budget aligned to your tolerance (e.g., $200 per month across a pilot org). Alert on 50/80/100% thresholds and rotate the budget monthly alongside your financial cadence. (docs.github.com)
3. Seat strategy: Use usage data to identify your heavy hitters and upgrade just those seats to Copilot Enterprise. Everyone else stays on Business. Re‑check quarterly. (docs.github.com)

Step‑by‑step: a 45‑minute Dec 2 audit (copy/paste this)

Run this once today, then repeat on the first business day each month.

1. Open Copilot → Spending. Screenshot current settings and export usage. (docs.github.com)
2. Set policy to Disabled unless you’ve consciously approved overages for a specific org.
3. If any org needs overages, create a Bundled budget with a clear owner and an amount that reflects reality (start small; raise with data). (docs.github.com)
4. Identify top 10 users by premium requests last month. For each, document model usage (look at multipliers) and whether Enterprise makes sense. (docs.github.com)
5. Communicate the change. Make it explicit that basic Copilot still works; premium features may pause when an allowance is hit. Link to internal guidance on which models to use by default. (github.blog)
6. Calendar a first‑of‑month review: counters reset at 00:00:00 UTC. Check 24–48 hours later to catch early spikes. (docs.github.com)

People also ask

Are Pro and Pro+ users affected by the Dec 2 budget removal?

No. GitHub has said Pro and Pro+ retain $0 budgets by default. The Dec 2 removal targets older Enterprise and Team accounts with an auto‑created $0 budget. (github.blog)

Will chat stop if we hit the allowance and block overages?

Chat with the included models continues on paid plans, subject to rate limits. Premium model interactions and features (agent, code review, Spark) are what stop. (github.blog)

How do model multipliers work in practice?

Each prompt consumes allowance multiplied by the model’s factor. For example, using a 10× model for one chat counts as 10 premium requests; using Gemini 2.5 Pro at 1× counts as one; using included models on paid plans uses zero. (docs.github.com)

A note on culture and defaults

Developers will default to whatever is fast and “just works.” If you silently block premium features without alternatives, they’ll feel it as friction. Instead, set a default model profile in IDEs that prioritizes included models for everyday tasks, and document when to switch up. Train teams on the agent’s sweet spots (PR scaffolding, repetitive edits) and reserve high‑multiplier models for thorny problems. The goal is not austerity; it’s predictable value.

Related reading from our team

If you need a deeper dive on settings and rollouts, we’ve published a step‑by‑step Dec 2 playbook, a breakdown of what’s changing on December 2, and practical advice to spend smart with Copilot. For help designing org‑level guardrails, see our AI and developer productivity services, or just talk to us.

What to do next (for today and the rest of December)

• Set your enterprise‑ or org‑level premium request policy to Disabled unless you’ve consciously approved overages. (docs.github.com)
• Create a small Bundled budget for any org that truly needs overages; assign an owner and alerts. (docs.github.com)
• Export usage, identify heavy users, and consider upgrading those seats to Enterprise if the math checks out. (docs.github.com)
• Publish a model usage guide: when to use included models vs premium, and how to request budget. (docs.github.com)
• Schedule a January 2 review—allowances reset monthly; check early patterns and tune budgets. (docs.github.com)

Zooming out: predictable AI costs beat reactive cuts

Usage‑based pricing isn’t going away. GitHub’s move toward policy‑driven controls and tool‑level SKUs is a nudge to treat AI as a shared utility with cost envelopes, not a magical flat‑rate. The teams who win set clear defaults, design guardrails, and tie premium consumption to outcomes (PR throughput, cycle time, defect trends). If you do that work now, Dec 2 is just another Tuesday.