GitHub Copilot users face sticker shock as usage-based pricing replaces flat fees

GitHub has officially transitioned its Copilot AI service from a request-based billing structure to a usage-based pricing model, a move that has triggered immediate and significant backlash from the developer community. The shift, which went into effect on 1 June 2026 following an April announcement, replaces the previous system where subscribers were allocated a set number of requests based on their payment tier. Under the new framework, users are granted monthly AI credits, with each credit valued at $0.01 of usage. This change aims to address the previous model’s inefficiency, where GitHub noted that a simple chat query and a multi-hour autonomous coding session incurred the same cost, forcing the company to absorb escalating inference expenses.

The financial impact on users has been stark, with many reporting what they describe as extreme sticker shock. Across social media platforms and technical forums, developers have shared data indicating that standard coding workflows now consume their monthly credit allotments at an alarming rate. Some users have reported depleting their entire monthly quota within a single day, a dramatic departure from the predictability of the former system. The cost of usage is now determined by the volume of input and output tokens processed, as well as the specific large language model selected for the task, leading to concerns over unpredictable and potentially exorbitant bills for complex coding tasks.

Pricing variability is a central feature of the new structure, with costs fluctuating significantly depending on the underlying model. For instance, one million output tokens from OpenAI’s GPT-5.4 nano cost $1.25, whereas the same output volume on the frontier GPT-5.5 model costs $30. Subscribers using the “Auto” mode, which selects the most appropriate model for a request, face additional risk, as some users report the system switching to expensive models for simple queries. Spot testing by Ars Technica demonstrated this variance, with a simple prompt to build a Minesweeper game consuming 94 credits via Claude Haiku 4.5, while other users reported single complex prompts burning through 171 credits or even 5,000 credits for Copilot-led commits.

Subscription tiers have been adjusted to reflect the new credit economy. The $10 per month Pro plan includes 1,500 credits, the $39 Pro+ plan offers 7,000 credits, and the $100 per month Copilot Max plan provides 20,000 credits. Despite these allocations, user sentiment remains largely negative. One developer noted that a cautious test of Claude Sonnet 4.6 consumed 840 credits, representing a significant portion of their monthly allowance without any complex work being performed. Another user reported that a single day’s activity accounted for 21 per cent of their Pro subscription’s total credits, leading to explicit threats to cancel their subscription or migrate to cheaper alternatives.

The reaction suggests a potential inflection point for the broader artificial intelligence market. While some developers are adapting by limiting usage to highly focused tasks and avoiding long-running chat sessions to conserve input tokens, others are exploring cost-effective alternatives. One user on Reddit highlighted the integration of Deepseek into their development environment, citing a cost of approximately seven cents for 15 million tokens. This shift in user behaviour may force competitors to reconsider their pricing strategies, potentially accelerating an industry-wide move toward usage-based models where efficiency in token usage becomes a key competitive advantage.