- GitHub Copilot switched from flat-rate Premium Request Units to token-based AI Credits billing on June 1, 2026
- Heavy agentic / reasoning model users report monthly costs jumping from $29 to $750+ — up to 25× increases
- Inline autocomplete remains unlimited and free on all plans; light users should see little to no change
GitHub quietly flipped the switch on its new billing model for Copilot on June 1, 2026. Out go Premium Request Units (PRUs) — the fixed multiplier system that let developers budget predictably. In comes AI Credits, a token-consumption model that aligns directly with the underlying API costs of whichever model you're using. The result: for developers leaning heavily on agentic workflows or high-tier reasoning models, the bill has become dramatically less predictable.
What Changed
Under the old system, each Copilot plan included a set number of "premium requests" per month — a bundled allowance before overage kicked in. Under AI Credits, charges accumulate based on actual input tokens, output tokens, and cached tokens consumed, priced according to each model's published API rates.
Base subscription prices remain unchanged:
| Plan | Monthly Price | Included Credits |
|---|---|---|
| Copilot Pro | $10/month | Limited AI Credit allowance |
| Copilot Pro+ | $39/month | Higher AI Credit allowance |
| Copilot Business | $19/user/month | Team-level allowance |
| Copilot Enterprise | $39/user/month | Enterprise-level allowance |
Overages beyond the included credits are billed at API rates for the model used — which is where the sticker shock comes in.
Developer Backlash
Reddit, X, and GitHub's own community discussion thread (#192948) filled rapidly with frustrated developers. TechCrunch summarized the sentiment with the headline "'What a joke': GitHub Copilot's new token-based billing spurs consternation among devs." Reports of bills jumping from $29 to $750/month, or from $50 to $3,000/month, spread widely in the first days of June.
The core complaint isn't the pricing model itself — it's the unpredictability. Agentic coding sessions where Copilot Agent runs multi-step workflows, fires off multiple reasoning model calls, and iterates on code can burn through credits in minutes without the user realizing how fast the meter is running.
The Other Side: Fairer for Light Users
Not everyone is unhappy. Developers who primarily use Copilot for inline completions and occasional chat — the majority of casual users — argue that the new system is fairer. Heavy Claude/GPT-4-class users now pay for what they consume, while light users retain predictable $10 Pro caps. GitHub's official position is that autocomplete usage, the feature most developers use most heavily, is still free and unlimited.
Migration Risk: Cursor, Windsurf Waiting
The billing change is accelerating consideration of alternatives. DEV Community published detailed cost guides comparing Copilot's new pricing against Cursor, Windsurf, and Cody. For teams doing heavy agentic coding, competing tools with flat monthly pricing now look more financially predictable — even if GitHub Copilot's IDE integration remains best-in-class.
Key Takeaways
- GitHub Copilot switched to token-based AI Credits billing June 1, replacing the PRU system
- Heavy agentic / reasoning model users report 10–25× monthly cost increases, some exceeding $750/mo
- Inline autocomplete remains unlimited and free — light users are unlikely to see meaningful cost changes
- Developer backlash has accelerated evaluation of alternatives like Cursor and Windsurf
- Pricing tables are now public and transparent, and base subscription prices did not increase