GitHub Copilot’s Usage-Based Billing Shift Draws Developer Pushback

GitHub moves Copilot toward token-based billing

GitHub announced that GitHub Copilot will move all plans to usage-based billing on June 1, 2026, with usage calculated through GitHub AI Credits rather than the current request-based system.

According to GitHub’s company blog, Copilot usage will be measured using input, output, and cached token consumption. GitHub’s pricing documentation says additional usage will be billed in AI Credits at per-token model rates, replacing the existing premium-request model.

TechCrunch reports that the change has prompted consternation among some developers, particularly those who rely on more complex Copilot workflows. The outlet described developer reaction to the new billing structure as frustrated, including concerns that multi-step and agentic coding sessions could become harder to predict or more expensive.

What is changing

Under the new system described by GitHub, Copilot usage will be tied more directly to the amount of model computation consumed. Instead of counting premium requests, GitHub says usage will be based on token consumption across model inputs, model outputs, and cached context.

GitHub Docs states that Copilot is moving from request-based billing to usage-based billing on June 1, 2026. The documentation also says additional usage will be billed through AI Credits, with different models carrying per-token rates.

That matters because coding assistants increasingly do more than complete a single line or answer a short prompt. In agentic coding workflows, tools may inspect files, generate plans, make multiple edits, run checks, and iterate on fixes. Each of those steps can add input and output tokens, and a long development session can involve repeated model calls.

Why developers are concerned

TechCrunch reports that the change directly affects developers who use costly multi-step Copilot and agentic coding workflows. Those workflows can consume more tokens than a simple chat exchange or inline code completion, making costs less transparent for users accustomed to a flatter subscription-like experience.

The concern is not just the existence of usage-based pricing, but predictability. A request-based allowance can be easier for developers to understand at a glance. Token-based accounting is more granular, but it can also be harder to estimate before a task is complete, especially when the tool decides how many steps to take.

GitHub’s own billing materials indicate that model choice will also matter, because additional usage is billed at per-token model rates. That means developers and organizations may need to pay closer attention to which Copilot features and models they use, particularly when working with larger codebases or longer prompts.

A broader shift in AI software pricing

GitHub’s move reflects a broader pattern in AI products: vendors are trying to align pricing with compute usage. Token-based billing is common among model providers because it maps more closely to the cost of processing prompts, generating responses, and maintaining context.

For individual developers, however, the practical question is whether the new structure makes Copilot feel less predictable. For teams, the issue may be budgeting and governance: administrators may need clearer controls, reporting, and model policies to keep usage within expected limits.

GitHub’s blog frames the transition as a move to usage-based billing across Copilot plans. TechCrunch’s report shows that at least some developers see the change as a meaningful departure from the simpler pricing experience they expected from Copilot.

What to watch next

The transition is scheduled for June 1, 2026, according to both GitHub’s announcement and GitHub Docs. Before then, developers and organizations using Copilot should review GitHub’s pricing reference, understand how AI Credits apply to their plans, and evaluate how token-heavy workflows may affect costs.

The biggest test for GitHub will be whether it can make usage transparent enough for developers to trust. If Copilot’s most advanced workflows are also the hardest to cost-estimate, the new billing model may remain a point of friction even if it more closely reflects underlying AI infrastructure costs.