Agentic AI: GitHub Copilot Reaches the Limits of Its Model

The token-based billing, an inevitable consequence for GitHub Copilot?

With the agentification of workflows, the pay-per-request business model does not appear tenable. GitHub is talking about it more and more openly. And it is taking steps. Officially, to “protect the experience of existing customers.”

Among these measures, there is the temporary halt on subscribing to new Pro, Pro+ and Student subscriptions. There is also a tightening of usage limits. GitHub does not quantify it, simply noting that the Pro+ plan includes five times more quota than the Pro plan. It has also decided to adjust the availability of certain models. At the top of the list is Claude Opus, which disappears from the Pro plan.

Read also: { Expert Opinion } – In 2026, the acceleration of software development will propel innovation

The agent-mode operation leads to long sessions and parallel processing. More and more users are exceeding the limits designed to keep the service available, we are told. Not without acknowledging that usage often incurs costs higher than the price of the plans…

These limits come in two forms. First, a weekly token volume limit, introduced recently in response to the mass adoption of agent-based requests. Second, a cap on the number of sessions within a time window that is not disclosed. Both are evolving over time.
The token limit does not prevent using the “premium requests” to which each subscription grants access. In the free version, it is 50 per month. On Pro and Pro+ subscriptions, it is 300 and 1500 respectively, with each additional request costing $0.04.

To minimize token consumption, GitHub recommends using the plan mode, which generates structured implementation plans before writing code. It also advises limiting the use of the /fleet command, which enables GitHub Copilot to create sub-agents.

GitHub had already ended trial periods

A few weeks ago, GitHub revisited its availability problems, which were becoming recurrent. Highlighting the “extreme growth” of usage on its platform, it acknowledged the “elasticity limits” of part of its infrastructure.

The company had notably promised to review its caching system and to “break the monolith” to better isolate key dependencies. It had also recalled that it is migrating to Azure, with the goal that half of its traffic passes through the US Central region.

At the beginning of April, GitHub had already partially suspended service access: it terminated all trial periods – including those in effect – for its Pro plan. Reason: an uptick in abusive usage.

More recently, the approach to telemetry has also changed. From now on, for individual subscriptions, GitHub collects interaction data by default (inputs, outputs, comments, documentation…) to train its models. Until now, it limited collection to interactions by employees of its parent company, Microsoft.
Default collection now extends to the GitHub Copilot CLI as well. With there also an option to opt out. Either via an environment variable (export GH_TELEMETRY=false or export DO_NOT_TRACK=true), or via a configuration option (gh config set telemetry disabled).

Dawn Liphardt

Dawn Liphardt

I'm Dawn Liphardt, the founder and lead writer of this publication. With a background in philosophy and a deep interest in the social impact of technology, I started this platform to explore how innovation shapes — and sometimes disrupts — the world we live in. My work focuses on critical, human-centered storytelling at the frontier of artificial intelligence and emerging tech.