GitHub Copilot Goes Token-Based: Devs Report 25x Bills
GitHub Copilot replaces flat-rate subscriptions with token-based billing on June 1, with some developers reporting costs jumping from $29 to $750 per month as agentic workflows drain credits in a single session.
7 min read
TL;DR
- GitHub Copilot replaces Premium Request Units with token-based AI Credits starting June 1, 2026
- Plan prices unchanged ($10/$39/mo), but now that money is a spend limit against API token rates
- Agentic coding sessions cost $30-40 each - a Pro user's monthly budget is gone in one sitting
- 893 negative reactions in the official community thread; Microsoft hasn't responded to press
- Code completions remain unlimited; every other feature is now metered
Today is the last day GitHub Copilot costs the same regardless of how much you use it.
Starting June 1, Microsoft's AI coding product drops its flat-rate model and moves to token-based billing, replacing a system of Premium Request Units with a new currency called GitHub AI Credits. The announcement has produced 893 negative reactions and more than 400 comments in the official community thread - numbers that don't usually show up when a company does something users like.
What Changes on June 1
From PRUs to AI Credits
GitHub Copilot's four plans keep their headline prices: Pro at $10 per month, Pro+ at $39, Business at $19 per user, and Enterprise at $39 per user. What changes is what you get for that money. Instead of a fixed pool of Premium Request Units, each plan now converts its subscription cost directly into AI Credits. One credit equals $0.01.
That means a Pro subscriber gets 1,000 credits per month. A Pro+ subscriber gets 3,900.
Those credits are consumed against published API token rates. Every input token, output token, and cached token in your interactions counts. Code completions and Next Edit Suggestions are carved out - they remain unlimited. Everything else, including Copilot Chat, agentic coding sessions, code review, and any use of premium models, draws from the credit pool.
The Model Pricing Table
| Model | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|
| GPT-5.5 | $5.00 | $30.00 |
| Claude Opus 4.5-4.8 | $5.00 | $25.00 |
| Claude Sonnet 4/4.5/4.6 | $3.00 | $15.00 |
| GPT-5.4 | $2.50 | $15.00 |
| Gemini 3.1 Pro | $2.00 | $12.00 |
| GPT-4.1 | $2.00 | $8.00 |
| Gemini 3.5 Flash | $1.50 | $9.00 |
| Claude Haiku 4.5 | $1.00 | $5.00 |
| GPT-5 mini | $0.25 | $2.00 |
Under the previous system, accessing Opus or GPT-5.5 cost the same fixed rate as using a cheaper model. Under the new one, it doesn't.
The Math Problem
One Agentic Session, One Month's Budget
GitHub has spent the past year marketing Copilot's agentic capabilities - the ability to plan, research, and execute multi-step coding tasks autonomously. Those workflows are exactly the ones that drain credits fastest.
Developers in the community thread calculated that a single agentic coding session with a premium model typically consumes between $30 and $40 in token credits. A Pro subscriber's entire monthly allotment is $10. A Pro+ subscriber gets $39 - enough for roughly one sizable agentic session, with nothing left for the rest of the month.
One community member noted that at Opus rates, a Pro+ plan supports about 140 completions before the credits run out - a number that experienced Copilot users described as roughly 90% fewer than what they could access before.
What Developers Are Saying
The reaction in the community thread is unambiguous. One developer wrote: "What a joke...This new usage model is just stupidly expensive. No longer cost-effective or useful in any practical way."
Another posted a screenshot showing projected monthly costs jumping from around $29 to roughly $750. A second user shared a similar estimate suggesting costs rising from $50 to around $3,000 - driven by heavy use of agentic workflows that GitHub's own product team had encouraged.
"Microsoft provided this billing method and they kept making it easier and easier to burn through massive numbers of tokens."
Not every developer is sympathetic to the outrage. Some have argued that costs at that scale only appear when someone is "purely vibe coding with a ton of bloated iterations" rather than using Copilot as a precision tool. Cursor, which this site covered in its Composer 2.5 release, operates a similar credit-based model at $20/month for Pro users, though its daily user caps have produced their own complaints.
Developer reactions in the official GitHub community thread. The post received 893 negative reactions within days of going live.
Source: github.com
Microsoft's Justification
GitHub's official explanation is that Copilot isn't the same product it was a year ago. The current version powers "far more complex, agentic workflows that consume far more compute," and the company frames the shift as necessary to deliver a "sustainable and reliable product experience" by aligning pricing to actual usage and costs.
That logic is accurate as far as it goes. Running extended agentic sessions with frontier models truly is more expensive than answering a single inline code question. What GitHub's explanation skips is any acknowledgment that the company spent a year actively pushing users toward exactly those expensive patterns - and that the previous pricing model absorbed those costs invisibly.
The fallback experience is also gone. Previously, when users exhausted their Premium Request Units, Copilot would downgrade requests to a cheaper model rather than stopping. Under the new system, when credits run out, access stops until the next billing cycle or until an admin purchases more.
Microsoft had not responded to TechCrunch's request for comment at the time of publication.
Who Gets Squeezed
Individual Developers and Small Shops
The new system has meaningful advantages for large organizations. Business and Enterprise accounts get pooled credits, budget controls, and three promotional months at enhanced limits: Business subscribers receive $30 in credits instead of $19 during June through August, and Enterprise subscribers get $70 instead of $39.
Individual developers on Pro plans don't get equivalent flexibility. A $10 monthly budget can't be increased through admin tools the same way an enterprise deployment can. Open source maintainers and solo developers - the users GitHub has historically cited as a core constituency - face the sharpest effective price increase relative to their previous access level.
Annual plan subscribers keep their PRU-based pricing until their plan expires, but model multipliers increase on June 1, which means even those users will see changes in how far their existing credits stretch.
The Alternatives Gaining Attention
The community thread is full of Cursor recommendations. At $20 flat for Pro, Cursor now competes with GitHub on AI coding features while maintaining a subscription model that doesn't require users to track token burn rates mid-session. Windsurf, also mentioned frequently, sits at $15 per month.
Cursor and Windsurf maintain flat subscription models while Copilot moves to token-based billing.
Source: techcrunch.com
GitHub is aware of the optics. The billing preview tool it launched in May gives users visibility into projected costs under the new system before the cutover arrives - though several developers noted that seeing the numbers early hasn't made them any happier.
The Pattern Here
The path of AI tool pricing follows a recognizable arc. Aggressive flat-rate subsidies draw in users and build dependency. Once usage patterns are established - and once the tools have shifted from nice-to-have to integrated into daily workflows - the pricing model adjusts to reflect actual costs.
GitHub's earlier moves were not subtle warnings. Copilot's data training policies generated controversy when default settings were changed without prominent notice. Ads appearing in pull requests were another departure from original expectations. Token-based billing is the third act in the same sequence.
None of that makes the per-session economics less jarring for developers who built agentic workflows around what they believed was an unlimited-model subscription. The tools that win from here may be the ones that kept their pricing model predictable - or the ones that are cheapest the day a developer cancels their Copilot subscription.
Sources:
