AI buffet is gone! GitHub Copilot's cost is too high to bear, starting from 6/1 it will be billed based on usage.

Unable to bear the soaring costs of compute power, GitHub Copilot announced that starting June 1, it will scrap its “unlimited” mode and switch to charging based on token usage. The move has triggered strong pushback from developers and a wave of subscription cancellations, highlighting the infrastructure challenges facing the AI industry.

The era of “unlimited AI” ends—GitHub Copilot changes its billing model

GitHub Copilot, once widely praised as “all-you-can-eat AI for the AI world,” is now turning from an all-you-can-eat buffet into a regular restaurant.

According to The Register, Microsoft-owned GitHub, in its latest announcement, admitted that due to an inability to continue absorbing losses, it has decided to start on June 1, 2026, switching the billing method from request-based to usage-based.

Under the original model, subscribers could submit a fixed number of advanced requests, without accounting for task complexity—resulting in prompt costs that require heavy computation far exceeding the subscription revenue received.

Image source: GitHub Copilot announcement of the end of the “unlimited AI” era, GitHub Copilot changes its billing model

GitHub product lead Mario Rodriguez also revealed that the cost of simple chats and AI code-writing tasks lasting several hours could be completely the same. The company has absorbed ever-rising inference costs, making the current model difficult to sustain.

As early as April 20, GitHub Copilot began adjusting its plans for individual subscribers, including pausing new user registrations for GitHub Copilot Pro, Pro+, and student plans, as well as tightening usage limits for individual plans, among other changes.

Starting June 1, GitHub Copilot introduces virtual pricing units

In the past, GitHub Copilot offered unlimited AI assistance services for a fixed monthly fee, which is why it was dubbed AI “all-you-can-eat,” and it has long been a low-key, high value-for-money option for engineers compared with mainstream solutions like Codex, Cursor, and Claude Code.

However, once it shifts to usage-based billing, GitHub Copilot’s charges will become directly tied to tokens.

Because different models have different fee rates, GitHub has created a virtual unit worth the equivalent of $0.01—GitHub AI points. Microsoft will convert the input, output, and cached tokens consumed by users into point costs based on the standard API fee rates.

Rodriguez said that going forward, GitHub Copilot’s subscription plans will include a fixed amount of AI points each month, and paid plans can also purchase additional points.

Because usage-based billing involves uncertainty, users cannot know in advance how many tokens a particular input will consume, and calculating it alongside other tools will be even more complicated. Therefore, GitHub plans to roll out a billing preview feature in early May, so users can assess their spending before the June transition period.

Reddit community backlash—netizens vow to cancel

Unsurprisingly, the changes to GitHub Copilot’s pricing have sparked a lot of backlash on the Reddit forum.

Some users responded that if charging is based on usage, there are already subscription-free services in the market like OpenRouter. The new system is essentially making users pay the full original API price, completely losing the value of a subscription.

Many annual subscribers feel their rights have been harmed. They estimate that the cost for certain models could jump by dozens of times, and they have all vowed to demand cancellations.

The community has also seen mounting calls to switch to other options. Many developers say they will use tools like Claude Code or Cursor instead, and some even choose to upgrade their hardware to run open-source models such as Alibaba’s Qwen 3.6 27B locally.

Image source: Reddit GitHub Copilot pricing changes spark backlash on Reddit

OpenClaw sparks a new craze—AI infrastructure overloaded

GitHub Copilot’s change reflects the broader infrastructure challenges faced by the AI industry.

Earlier this year in February, the open-source AI assistant OpenClaw—nicknamed “lobster”—drew massive attention, prompting many developers to experiment with AI agents running tasks around the clock, 24 hours a day. Improvements in model capabilities have also encouraged more people to explore AI coding.

This has led AI companies that previously offered subscription subsidies to face an enormous demand far exceeding what their inference infrastructure can handle. Even AI giants such as Anthropic and OpenAI have encountered capacity problems. Claude Code only recently fixed a major bug, improving output quality and addressing issues with poor performance and latency (commonly referred to as getting “dumber” / “less intelligent”), and it also reset users’ usage quotas.

Until the industry finds a way to balance costs and user experience, the resource consumption driven by massive AI compute demand—and its “price correction effect”—will continue to spread across the entire AI industry.

Further reading:
Claude Code got dumber for real! The official admits there are 3 major bugs—users’ subscription quotas are reset in full

Lawmakers propose banning data centers; environmental groups criticize an ecological disaster—meanwhile, the First Lady is walking into the White House with an AI robot