GitHub flipped the switch today. Copilot's flat-rate subscription is gone. In its place: a token meter that's turning $29 monthly bills into $750 invoices before developers have finished their morning standup.
On Reddit, the rage is specific. One Pro user's $29 plan projected to $750. Another saw a jump from $50 to $3,000. Hundreds of threads across HN, Reddit, and X are all asking the same question: how does a developer tool get more expensive than an AWS account?
The Numbers
Let's be concrete. Under the old plan, Copilot Pro was $10/month for code completions plus chat. Pro+ was $39 for the premium models. Everything after that was "premium requests" with vague limits that most people never hit.
Here's what changed on June 1:
| Plan | Old Price | New Monthly Credits | What Credits Buy |
|---|---|---|---|
| Pro | $10/user | ~1,000 credits ($10) | ~10 hours of chat-only |
| Pro+ | $39/user | ~3,900 credits ($39) | ~4 hours of agentic work |
| Business | $19/user | ~1,900 credits ($19) | ~8 hours of mixed use |
| Enterprise | $39/user | ~3,900 credits ($39) | Depends entirely on agent use |
One credit equals one cent. Code completions stay free. Everything else is metered: Copilot Chat, the CLI, cloud agents, automated code reviews, Copilot Spaces, Spark, and third-party agents. Code reviews get billed twice: once for the token credits, once for GitHub Actions minutes.
A Pro user burning through chat and agents can exhaust their monthly credit pool in about 36 hours of active work. That's less than a standard work week.
GitHub did offer a cushion. Business users get an extra $30/user in credits through August. Enterprise users get $70. After September, you're on the 1:1 ratio. Your subscription price equals your credit pool, and everything beyond that costs extra.
Community Reaction
The developer response has been visceral. A TechCrunch story published on May 30 collected reactions from across the community:
"The only way it gets crazy like that is if you are purely vibe coding with a ton of bloated iterations," one Reddit user wrote, defending the pricing model against accusations of gouging.
Others pointed the finger at Microsoft itself. "Microsoft provided this billing method and they kept making it easier and easier to burn through massive numbers of tokens on single premium requests that could churn for hours or even days while spawning dozens or even hundreds of sub-agents."
A more cynical take from a different thread: "Holy fuck how much money was copilot losing?"
The broader sentiment was summarized neatly by one developer on X: "You will get less, but pay the same price."
On Hacker News, the discussion has been more analytical. Several commenters noted that the shift mirrors what Cursor did in June 2025 and Windsurf did in March 2026. The industry is converging on a single model, and the flat-rate era is ending across the board.
The Competitive Landscape
This doesn't happen in a vacuum. Here's what the alternatives look like right now:
| Tool | Pricing Model | Monthly Cost (typical) |
|---|---|---|
| Cursor | Usage-based since June 2025 | $16-20/mo (Pro, limited usage) |
| Claude Code | $17/mo Pro + metered compute | $17-200/mo depending on usage |
| Windsurf | Free for individuals | $0-15/mo |
| Cline (OSS) | Bring your own API key | Pay per token, typically $10-50/mo |
| Aider (OSS) | Bring your own API key | Pay per token, typically $10-30/mo |
The open-source alternatives look increasingly attractive. Aider and Cline let you use any model through your own API key, which means you can use cheaper providers like OpenRouter or DeepSeek directly. The trade-off is setup time and missing enterprise features like SOC 2 compliance and JetBrains integration.
Cursor has been on usage-based billing for a year and has reached $500M+ ARR. The market has already voted that this model works, at least for the vendors.
Microsoft's Quiet Response
Two days after the billing announcement, VS Code 1.118 shipped with three specific features: deferred tool loading, smarter prompt caching, and smaller default models. Together, they claim up to 20% token savings.
The timing is not coincidental. Microsoft knows the first bills will sting. The VS Code updates are a damage-control measure that reduces token consumption so the June sticker shock is slightly less traumatic.
They also halted new Copilot signups earlier this year, citing "soaring usage and rising costs." The flat-rate model was described internally as "currently unsustainable." This change is a financial corrective that transfers the cost risk from GitHub to the people actually using the tool.
What This Actually Means
The flat-rate AI era is over. Not just for Copilot but for everything. Cursor ended it last year. Windsurf ended it this spring. Now Microsoft has made it official: if you want AI in your editor, you will pay per token, and the meter runs faster than you think.
The immediate winners are the open-source tools. Cline and Aider don't have shareholders to satisfy. They pass API costs through at cost, or close to it. The loser is the idea that AI coding assistance can be a flat-rate utility like electricity. It turns out the marginal cost of generation is real, and the market has decided someone has to pay it.
If you use Copilot casually, sticking to completions only without chat or agents, your bill probably stays flat. Code completions aren't metered. But if you've been using Copilot Chat, if you've tried Agent Mode, if you've automated code reviews, you are about to get a very specific number telling you how much GitHub thinks your workflow costs.
The number might make you reconsider the open-source path. It might make you optimize your prompts. Or it might just make you angry. Based on the threads today, most people are somewhere between angry and confused, and I don't blame them.
Nobody likes finding out the price after they've already bought the product.