Customers revolt as GitHub Copilot 'fixes' rate limits

The change

GitHub told Copilot customers last week they must rein in usage as the company works to “better balance capacity” and improve service reliability. It has been reported that the move follows discovery of a bug that undercounted token usage from newer models — a flaw that allegedly broke parts of GitHub’s pricing calculus and let consumption run far higher than expected. To cope, GitHub is imposing stricter rate limits, retiring Anthropic’s Opus 4.6 Fast for Copilot Pro+ users, and suspending all Copilot Pro free trials.

Backlash

Developers are not happy. Users report short, repeated interruptions and, worse, multi-day or even 44-hour “weekly rate limits” that halt work cold. One Copilot Pro Plus hobbyist told The Register he was forced into Auto mode — supposedly a fallback — and suddenly the assistant started taking shortcuts and producing worse output. Community threads have exploded with complaints about “obscenely long rate limits,” and the emotional tenor is unmistakable: frustration, lost momentum, and a feeling of being left in the lurch.

Bigger picture

GitHub frames this as a capacity and reliability issue. But it has been reported that the underlying problem is financial: Roman Kir of StratoAtlas wrote that in March GitHub discovered its rate-limiter had been undercounting tokens from models like Claude Opus 4.6 and GPT-5.4, driving unanticipated costs. Who pays when an “all-you-can-eat” token buffet runs wild — the company, the users, or the models themselves? Expect more tightrope-walking as platform vendors and model providers juggle demand, costs, and developer trust.

Sources: The Register