GitHub switched its Copilot AI coding assistant to usage-based pricing on Monday, replacing flat-rate request limits with monthly credit allotments. Some users exhausted their entire monthly allowance in under 24 hours, according to posts on social media and forums. One developer reported burning through 840 credits on the first day while being 'super cautious.'
The new system assigns a credit value to each AI interaction based on token usage and model choice. A single complex prompt consumed 171 credits, one user reported. Another spent 700 credits on 'a few prompts.' Ars Technica confirmed these figures in spot testing published Monday.
GitHub's Pro plan now includes 1,500 credits per month for $10. That is $15 worth of usage. The Pro+ tier offers 7,000 credits for $39.
The Max plan provides 20,000 credits for $100. Credits cost one cent each. The math is simple.
But the underlying costs are not. Different models charge drastically different rates. One million output tokens from OpenAI's GPT-5.4 nano cost $1.25 on Copilot.
The same output from GPT-5.5 costs $30. Users relying on 'Auto' mode may not realize which model is selected. Some report the system picks expensive models for trivial queries. 'Even though I was super cautious on the first day, trying it out with a limited number of uses, it still consumed 840 credits,' one user wrote of testing Claude Sonnet 4.6 through Copilot.
Another complained about using 21 percent of a monthly Pro allotment in a single day. 'I haven't even done any really complex work yet.'
GitHub defended the change in April. The old system treated a quick chat question and a multi-hour autonomous coding session as equal. That forced Copilot to absorb escalating inference costs.
The company said the new model aligns price with value. Some users are adapting. Coder Henri Kinnunen reported using only 161 credits in a productive day with GPT 5.3-Codex.
The secret: 'very focused and deliberate changes with AI.' Neil Hewitt noted on Bluesky that continuing a three-day-old chat session now sends the entire history as context each time. 'Input tokens use credits... it's not rocket science.'
Others are looking elsewhere. Reddit users discuss integrating Deepseek into VS Code at about 7 cents per 15 million tokens. That is a fraction of Copilot's rates.
But cheaper models may deliver poorer results. 'You get what you pay for,' one commenter noted. Now users must also pay for what they get. Here is what they are not telling you.
The credit burn rate is not just about model choice. It is about context length. Every time a developer continues an old chat, the entire conversation history is sent as input.
That consumes credits fast. The math does not add up for users who treat Copilot as a persistent coding partner. The pricing shift reflects a broader industry trend.
AI companies are moving from subsidized user acquisition to sustainable unit economics. OpenAI, Anthropic, and Google have all raised prices or introduced usage limits. Venture capital is no longer covering inference costs.
Follow the leverage, not the rhetoric. GitHub's move may accelerate a shakeout among AI coding tools. Services that offer generous free tiers or flat-rate pricing could gain users.
But those services may face the same cost pressures. The question is whether developers will pay premium prices for frontier models or switch to cheaper, less capable alternatives. The economic stakes are high.
GitHub Copilot has over 1.8 million paid subscribers, according to Microsoft's last earnings call. Even a small churn rate could mean millions in lost revenue. But the cost of serving power users may be even higher.
One estimate shared by a user showed their previous monthly usage would cost over $2,000 under the new plan. That timeline concerns investors. Microsoft has bet heavily on AI integration across its products.
Copilot is a flagship offering. If usage-based pricing drives users away, the narrative around AI monetization could shift. Enterprise customers may demand fixed-rate contracts.
Individual developers may defect to open-source alternatives. The backlash is not universal. Some users welcome the transparency.
They prefer paying for what they use rather than subsidizing heavy users. 'I'd rather pay for my own usage than have limits imposed because of someone else's,' one Reddit user wrote. The debate mirrors earlier fights over internet data caps and cloud computing pricing. Historical parallels exist.
In the early 2000s, ISPs moved from unlimited dial-up to metered broadband. Users revolted. Then they adapted.
The same pattern played out with cloud services. AWS launched with per-hour billing. Customers demanded reserved instances and savings plans.
GitHub may need to offer similar flexibility. The key question is whether GitHub adjusts its credit allotments or pricing tiers. The company has not commented on the backlash.
But user forums are filling with cancellation threats. Some have already switched to competitors like Cursor, Codeium, or Amazon CodeWhisperer. Those services may see a surge in sign-ups this week.
The competitive landscape is shifting. Cursor offers a $20 pro plan with 500 fast premium requests. Codeium's individual plan is free.
Amazon CodeWhisperer is free for individual use. These alternatives may not match Copilot's deep GitHub integration. But for many developers, price trumps convenience.
The developer experience is changing. AI coding assistants were sold as productivity boosters. Now they are becoming cost centers.
Teams will need to budget for AI credits like they budget for cloud compute. That could slow adoption among freelancers and small startups. It may also encourage more selective use of AI.
One vivid detail: a developer posted a screenshot showing a single commit consumed 2,500 credits. The commit message was 'fix typo.' The irony was not lost on the community. 'I could have fixed that myself in 10 seconds,' the user wrote. Why It Matters: The Copilot pricing change is a bellwether for the AI industry.
If developers reject usage-based pricing, it could force a rethink of how AI tools are monetized. If they accept it, expect similar moves from every AI-powered productivity tool. The outcome will shape the economics of AI for years.
The broader significance extends beyond coding. AI is being embedded into email, documents, and design tools. Flat-rate pricing encourages overuse.
Usage-based pricing risks alienating users. The industry is watching GitHub's experiment closely. What comes next is a period of experimentation.
GitHub may introduce rollover credits or family plans. Competitors may seize the moment with aggressive pricing. Enterprise customers will demand predictability.
The market will sort out winners and losers over the next six months. Key takeaways: - GitHub Copilot's new usage-based pricing burns through monthly credit allotments in hours for some users, sparking cancellation threats. - Credit consumption varies wildly by model choice and context length, with frontier models costing 24 times more than nano models. - The shift reflects an industry-wide move from subsidized growth to sustainable pricing, with implications for all AI productivity tools. Developers should monitor their credit usage closely.
The 'Auto' model selection can be costly. Starting fresh chats instead of continuing old ones saves credits. And the landscape will look different by year's end.
Some AI coding tools will not survive the transition to usage-based pricing. The ones that do will have solved the hardest problem in AI: making money.
Key Takeaways
— - GitHub Copilot's new usage-based pricing burns through monthly credit allotments in hours for some users, sparking cancellation threats.
— - Credit consumption varies wildly by model choice and context length, with frontier models costing 24 times more than nano models.
— - The shift reflects an industry-wide move from subsidized growth to sustainable pricing, with implications for all AI productivity tools.
Source: Ars Technica
