GitHub Copilot Customers Revolt Over Aggressive Rate Limits Following Token Counting Bug Fix
Key Takeaways
- ▸GitHub discovered a critical token counting bug that underestimated usage from newer AI models, masking true infrastructure costs and breaking the company's pricing model
- ▸Newly implemented rate limits are causing customer backlash, with some users facing 44-hour to multi-day restrictions despite paying premium subscriptions
- ▸GitHub has also suspended Copilot Pro free trials due to abuse and retired Anthropic's Opus 4.6 Fast model for Pro+ users as cost-control measures
Summary
Microsoft's GitHub has implemented strict rate limits on Copilot users following the discovery of a token counting bug that significantly undercounted usage from newer AI models like Claude Opus 4.6 and GPT-5.4. The bug masked the true infrastructure costs, leading to unexpectedly high consumption that strained company servers. Customers now report facing rate limits ranging from hours to multiple days, with some premium users spending hundreds of pounds monthly on credits finding their service access severely restricted. The situation has prompted widespread complaints in GitHub Copilot community forums, with users expressing frustration over the sudden shift from an all-you-can-eat pricing model to aggressive throttling that appears designed to control unexpected costs.
- Similar capacity and cost management issues are affecting competitors including Anthropic and OpenAI's Codex, indicating broader industry challenges with AI service economics
Editorial Opinion
GitHub's handling of its token counting bug reveals the precarious economics underlying AI-as-a-service models when pricing assumptions break down. While infrastructure strain is legitimate, the aggressive and seemingly opaque rate limiting—with some users facing 44-hour lockouts—risks eroding customer trust and suggests the company struggled to communicate transparently about the underlying cost problem. This situation underscores a critical industry lesson: as AI models become more capable and expensive to run, all-you-can-eat pricing models are unsustainable without robust usage monitoring and honest customer communication from day one.
