GitHub Copilot's New Pricing: The Real Cost of AI-Assisted Coding
GitHub's Copilot usage-based pricing launched today, and the numbers are shocking some developers. Under the new system, paid subscriptions grant a monthly allotment of AI credits (1 credit = $0.01). The $10/month Pro plan gives 1,500 credits ($15 worth), the $39 Pro+ plan gives 7,000 credits ($70 worth), and the $100/month Copilot Max plan gives 20,000 credits ($200 worth).
How Credits Burn
Credit consumption depends on model and token count. One million output tokens from OpenAI's GPT-5.4 nano cost $1.25, but the same output from GPT-5.5 costs $30. Users relying on "Auto" mode risk expensive surprises—reports show it can switch to costly models for simple queries.
Ars Technica tested a simple "build a Minesweeper game" prompt via Claude Haiku 4.5, costing 94 credits. A single complex prompt reportedly burned 171 credits, while "a few prompts" cost 700 credits. One user saw 5,000 credits consumed in a couple of Copilot-led commits. Even a "run-of-the-mill query" cost 15 credits, and "generating a small plan" used 100 credits.
Real User Pain
One developer testing Claude Sonnet 4.6 through Copilot wrote: "Even though I was super cautious on the first day, trying it out with a limited number of uses, it still consumed 840 credits." Another reported using 21% of their monthly Pro subscription allotment in a single day, complaining: "I haven't even done any really complex work yet." A user on X showed an org's entire 8,000 monthly credits burned in one day.
Adjusting or Jumping Ship
Some developers are adapting. Henri Kinnunen reported only 161 credits in a "productive day" using GPT 5.3-Codex, achieved by "very focused and deliberate changes with AI." Neil Hewitt noted on Bluesky that continuing a three-day-old chat session is unwise now because "the entire chat history as context every time... input tokens use credits."
Others are leaving. One Reddit user integrated Deepseek into VS Code for "about 7 cents for 15 million tokens." As GitHub's subsidized pricing fades, industry-wide usage-based models may follow, making token-efficient LLMs the economic winners.
What This Means for Developers
- Monitor your credit usage with GitHub's dashboard.
- Avoid long-running chat sessions—they burn input tokens.
- Consider switching to cheaper models (e.g., GPT-5.4 nano) for routine tasks.
- Explore alternatives like Deepseek if your usage is heavy.
The era of unlimited AI coding assistance is over. Plan your prompts like you're paying per token—because you are.
