The $10-a-month AI coding assistant is becoming a $10-a-month gateway to a much bigger bill.
Leaked internal documents obtained by Ed Zitron’s Where’s Your Ed At reveal that Microsoft is preparing to shift GitHub Copilot from flat-rate subscription pricing to token-based billing — charging users for the actual compute their prompts consume. The weekly cost of running GitHub Copilot has nearly doubled since January, and the current pricing model cannot sustain it.
The change is coming for every tier. No one gets to keep the deal they signed up for.
What Is Changing
Three moves, all designed to cut Microsoft’s losses:
1. Token-based billing replaces requests
GitHub Copilot currently uses a “request” system — each interaction with the AI counts as one or more requests, depending on the model. Pro accounts ($10/month) get 300 requests. Pro+ ($39/month) gets 1,500.
Token-based billing means you pay for what you actually use — input tokens (your prompt and context) and output tokens (the model’s response, including chain-of-thought reasoning). Claude Opus 4.7 costs $5 per million input tokens and $25 per million output tokens. A heavy coding session can burn through that quickly.
Microsoft has confirmed this is coming later in 2026. The documents say it became a top priority as costs spiralled.
2. Opus removed from the cheap tier
Anthropic’s Opus family — the most capable and most expensive models — is being pulled from the $10/month Copilot Pro plan entirely. Opus 4.6 Fast was already retired in early April. Opus 4.6 and 4.5 will disappear from Pro+ in the coming weeks, replaced by Opus 4.7.
The catch: Opus 4.7 has a 7.5x request multiplier — each prompt counts as 7.5 requests. The previous Opus 4.6 had a 3x multiplier. That means Opus 4.7 is roughly 250% more expensive per prompt than its predecessor, even before token billing kicks in.
Microsoft is offering the 7.5x rate as a promotional discount until April 30. What happens after that is unclear, but the trajectory is obvious: the best models will cost substantially more to access.
3. Rate limits tightened across the board
Individual, Business, and Enterprise plans are all getting tighter rate limits. Microsoft already reduced limits at the start of April. The documents say it was not enough, and further reductions are coming within weeks.
Signups for individual and student tiers will be temporarily paused. Paid plan trials will be suspended to “fight abuse” — which in practice means preventing users from cycling through free trials to access subsidised compute.
Why This Matters
Microsoft is not alone. Anthropic recently shifted enterprise users to token-based billing. OpenAI has been steadily increasing prices for API access. Cursor, the AI coding startup, has already moved to usage-based pricing.
The pattern is clear: the AI industry subsidised user acquisition by offering frontier models at below-cost prices. That strategy worked for growth. It does not work for sustainability.
Microsoft’s own numbers tell the story. Weekly costs for running GitHub Copilot have doubled since January. Every user generating code with Opus is burning compute that costs multiples of what their subscription covers. The $10/month Pro plan that includes Opus access has been, from Microsoft’s perspective, a loss leader.
The subsidy era is ending. The question is whether users who got hooked on cheap AI will pay what it actually costs.
What This Means for Developers
For individual developers on Copilot Pro ($10/month): You are losing access to the most capable models. Your remaining models (GPT-5.4 Mini at 0.33x, Sonnet at 1x) still work, but the quality gap between what you get for $10 and what the top models can do is about to widen significantly.
For Pro+ users ($39/month): Opus 4.7 costs 7.5x per request. Your 1,500 monthly requests translate to roughly 200 Opus prompts before you hit the wall. That is a heavy coding day, not a month.
For Business and Enterprise: Rate limits are tightening. Token billing is coming. Expect your AI coding bill to increase — potentially significantly — as you pay for actual usage rather than a flat seat.
For the AI-curious: If you have been considering subscribing to an AI coding tool, the calculus has changed. The flat-rate deals that made these tools no-brainers are disappearing. The new pricing rewards efficiency — short prompts, smaller models, focused use cases — and punishes the exploratory, conversational coding style that made AI assistants feel magical.
The Bigger Picture
Every major AI company is making the same move for the same reason: the compute does not pay for itself.
This is the second wave of AI economics. The first wave was land grab — get users at any cost, subsidise the product, grow fast. The second wave is unit economics — make the product pay for itself, charge what it costs, optimise for margin.
For workers, this means the tools that were becoming standard issue are getting more expensive and more restricted. The companies that gave away AI to build habit are now extracting payment. The question is not whether this will happen to every AI product. It already is. The question is whether the tools are valuable enough that people will pay the real price.
Based on the numbers Microsoft is seeing — costs doubling while usage stays high — the answer appears to be yes. People will pay. They just will not be happy about it.
Sources
- Ed Zitron, Where’s Your Ed At — “Microsoft To Shift GitHub Copilot Users To Token-Based Billing”
- GitHub Blog — Changes to GitHub Copilot Plans (April 2026)
- GitHub Blog — Claude Opus 4.7 General Availability
- The Information — Anthropic shifts enterprise users to token-based billing
