GitHub said long-running, parallelized AI coding sessions are pushing Copilot beyond the limits of its original individual plan structure, prompting tighter caps and a pause on new sign-ups.
GitHub has paused new sign-ups for several individual Copilot plans and tightened usage limits, saying newer agentic coding workflows are consuming far more compute than its original pricing and service model was built to handle.
The move is a reminder that as AI coding assistants grow more autonomous, vendors may have to balance developer demand against infrastructure cost and service reliability.
“As Copilot’s agentic capabilities have expanded rapidly, agents are doing more work, and more customers are hitting usage limits designed to maintain service reliability,” GitHub said in a blog post. “Without further action, service quality degrades for everyone.”
Under the changes, GitHub has paused new sign-ups for its Copilot Pro, Pro+, and Student plans, saying the move will help it better serve existing customers.
The company is also tightening usage limits on individual plans, while positioning Pro+ as the higher-capacity tier with more than five times the limits of Pro for users who need heavier usage.
At the same time, GitHub is narrowing model access: Opus models will no longer be available on Pro plans, while Opus 4.7 will remain on Pro+, and Opus 4.5 and 4.6 are also set to be removed from that tier.
GitHub said it will now show usage limits directly in VS Code and Copilot CLI so users can more easily track how close they are to those caps.
The company added that affected Pro and Pro+ users who contact support between April 20 and May 20 can request a refund and will not be charged for April usage if the updated plans do not meet their needs.
GitHub’s move comes as other AI vendors are also adjusting usage policies to manage capacity, with Anthropic last month changing how Claude’s timed limits work during peak hours while keeping weekly limits unchanged.
Charlie Dai, vice president and principal analyst at Forrester, said the move shows how agent-driven coding is shifting workloads toward longer-running and parallel sessions that create higher and less predictable compute demand.
“Cost structures built for lightweight assistance no longer hold, and this puts pressure on GPU capacity, reliability, and unit economics,” Dai said.
Dai added that similar usage restrictions by major model providers suggest capacity rationing is likely to become a structural feature of the industry as agentic development becomes more routine.
Impact for developers
GitHub said Copilot now operates with both session limits and weekly seven-day limits, and that those caps are based on token consumption and model multipliers rather than just raw request counts. Users may still have premium requests left and yet hit a usage limit, because the two systems are separate.
In practice, that means developers using heavier agent-style workflows, especially long-running or parallel sessions, are more likely to hit limits than those using Copilot for simpler tasks.
GitHub is encouraging users nearing their caps to switch to lower-multiplier models, use plan mode in VS Code and Copilot CLI, and cut back on parallel workflows such as /fleet.
Analysts said the move also reflects a familiar pattern in the tech industry.
“First you give users access to a tool with relatively open usage, and then gradually start defining limits as adoption grows,” said Faisal Kawoosa, founder and chief analyst at Techarc. “GitHub has an unavoidable role in the developer world. A developer can live without an email ID, but not a GitHub account. Such is the depth of its integration. But at the same time, the rationalization of AI/Copilot in the ecosystem is inevitable, as resources are constrained.”
Kawoosa added that developers have now seen what Copilot can do, and there is little reason for GitHub to keep offering it without tighter limits. He said the next step is likely to be more differentiated plans that create clearer monetization opportunities among individual users. For enterprise engineering leaders, Dai said the episode is a reminder to evaluate AI coding tools as metered infrastructure rather than unlimited productivity layers. He said buyers should pay close attention to usage ceilings, downgrade behavior, model entitlements, and how clearly vendors communicate limits and cost controls to developers.
