GitHub Copilot Hit a Limit Its $10-a-Month Pricing Model Never Saw Coming

GitHub's AI Coding Assistant Hit a Limit Its Pricing Model Never Saw Coming

On April 14, a developer identified only as John Clary received a notification that many Copilot users have learned to dread: his access had been cut off. He was 44 hours into what GitHub calls a weekly rate limit, a hard ceiling on how many tokens he could burn through in a seven-day window. The work he had been doing stopped mid-session. There was no downgrade option, no escalation path, no explanation beyond a number on a screen.

"I use GitHub Copilot Pro Plus as a hobbyist and frequently spend hundreds of pounds a month on additional credits," Clary told The Register. Switching to the cheaper Auto mode, where Copilot picks the cheapest model, brought the service back. It also brought the performance down far enough that he described it as frequently taking shortcuts without telling him.

Clary was not unlucky. He was a preview of what GitHub had been quietly discovering for weeks: the subscription he was paying for no longer bore any relationship to what it cost GitHub to serve him.

On April 20, GitHub confirmed what a growing number of locked-out users had already figured out. The company paused new sign-ups for Copilot Pro, Copilot Pro+, and Student plans, effective immediately. The official reason was service reliability. The real reason, buried in technical notices and internal documents reviewed by Ed Zitron's Where's Your Ed At, is that GitHub's pricing model had collapsed.

The trigger was a bug. In March 2026, GitHub discovered that its rate-limiting system had been systematically undercounting token consumption from two of its most popular premium models: Anthropic's Claude Opus 4.6 and OpenAI's GPT-5.4. Both models consume significantly more infrastructure per request than their predecessors, more tokens in, more tokens out, longer chains of reasoning. But the counter measuring that consumption had been wrong. When GitHub fixed the bug, limits that had been calibrated to a false reading snapped back to their intended values. Users who had been operating normally under the old broken meter suddenly found themselves facing 181-hour lockouts. Coding agents were killed mid-task. There was no recovery path and no upgrade tier above Pro+.

The bug revealed a structural problem that had been accumulating since the moment agentic workflows went mainstream. "The unit of sale, a subscription, a plan tier, had been decoupled from the unit of actual cost," wrote Roman Kir, founder of research consultancy StratoAtlas, in an analysis of the incident. Once GitHub corrected the measurement, the price of serving its heaviest users became visible. It was higher than the subscriptions were designed to cover.

Internal documents viewed by Where's Your Ed At show that the weekly cost of running GitHub Copilot nearly doubled between January and April 2026. The flat-rate subscription, $10 per month for Copilot Pro and $39 per month for Pro+, was built around a world where Copilot was an autocomplete tool for individual developers. It was not built for autonomous agents running hundreds of requests in parallel, burning through tokens at rates that multiply with model quality. The premium models developers want most are also the ones that cost the most to run, and they cost far more per session than the subscription model assumes.

"The cost structures built for lightweight assistance no longer hold," said Charlie Dai, vice president and principal analyst at Forrester Research. "This puts pressure on GPU capacity, reliability, and unit economics." Dai's verdict: the economics of a $10-per-month flat-rate AI coding tool, when actual compute burn from agentic workflows can exceed what the subscription covers, is broken at the foundation.

GitHub's response has been to do what any utility does when demand outstrips supply: raise prices and ration access. It is tightening usage limits across individual plans. Opus models are being removed from lower tiers, Opus 4.5 and 4.6 will be gone from Pro+ within weeks, with the newer Opus 4.7 reserved for higher tiers at a premium multiplier. And the company is moving toward token-based billing, where users pay for what they actually burn rather than a fixed bundle of requests. That transition will raise costs for most individual subscribers.

GitHub has offered refunds to Pro and Pro+ users unhappy with the new limits. They have until May 20 to request one and will not be charged for April. But the refund window is an acknowledgment of the problem, not a solution to it. The underlying issue is that the flat-rate subscription era for AI coding tools was subsidized on the assumption that usage would stay within bounds that a monthly price could cover. That assumption broke when agentic workflows crossed from experiment to production.

"The rationalization of AI and Copilot in the ecosystem is inevitable, as resources are constrained," 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 account, but not a GitHub account, which means its pricing decisions ripple across the entire ecosystem.

For individual developers, the practical consequences are concrete: the tools they built workflows around are more expensive, more limited, or temporarily unavailable. For the industry, the GitHub episode is a case study in what happens when the unit economics of AI assistance meet the actual usage patterns of autonomous agents. Every AI-as-subscription product faces the same math. The subsidized flat-rate era had a good run. The bill has arrived.

GitHub declined to comment for this article.