[Digital Today reporter Jinju Hong (홍진주)] Backlash is growing among GitHub Copilot users over a new pay-as-you-go billing system.

On June 1 local time, IT outlet Ars Technica reported that users have posted repeated cases saying the AI credit-based pricing plan that took effect that day quickly depleted monthly limits even with normal usage.

The change was announced in April. Previously, each subscription tier came with a set number of requests and premium requests. Now users receive monthly AI credits that are deducted based on actual usage. One credit corresponds to $0.01 of usage.

Each paid subscription now has a set included amount. The $10-a-month Pro plan provides 1,500 credits, the $39-a-month Pro+ provides 7,000 credits, and the $100-a-month Copilot Max provides 20,000 credits. The problem is that even within Copilot, costs vary widely depending on which large language model (LLM) is used. Deducted credits differ depending on the number of input and output tokens and the base model price.

GitHub said that under the previous system, "a simple chat question and a multi-hour autonomous coding session were processed at the same cost to the user" and that the service had been absorbing a large share of rising inference costs. Users, however, say the new system does not match real usage patterns. Some claimed that converting their prior month of usage to the new pricing, based on GitHub's own calculator tool, would raise costs to several thousand dollars.

Examples of actual use are also spreading quickly. One report said a single complex prompt consumed 171 credits, and another said "a few prompts" used 700 credits. One user said about 5,000 credits were deducted for a couple of Copilot-led commits. Others said a simple query cost 15 credits and generating a small plan took 100 credits.

One user who tested Claude Sonnet 4.6 wrote, "Even though it was the first day so I used it very carefully and in a limited way, 840 credits were consumed." Another user said, "Just one day of use spent 21 percent of the Pro plan monthly limit," adding, "I haven't done any really complex work yet." The user added, "I think I'll be going elsewhere soon."

Some developers, on the other hand, said changing how they use it can reduce the burden. Developer Henri Kinnunen said he used only 161 credits during a productive day using Claude 5.3-Codex in Copilot. He said he limited AI use to "very focused and deliberate changes." Developer Neil Hewitt also said that continuing a conversation from three days earlier in Copilot is now inefficient, noting that "the entire conversation history gets fed back in as context every time."

Users also point to model-by-model cost differences as a factor that increases the burden. In Copilot, 1 million output tokens from OpenAI GPT-5.4 Nano costs about $1.25, but the GPT-5.5 model costs $30 for the same output size. Users also reported that when using Copilot's auto-selection mode, it can switch to a more expensive model even for simple questions.

Ars Technica reran a simple prompt to create a Minesweeper game using Claude Haiku 4.5 and found it consumed about 94 credits. It may look manageable for a simple sample project, but it showed that costs can rise sharply when moving to complex tasks such as modifying or reviewing large codebases.

Some users are moving to other AI coding services that offer more generous usage limits. There is also a possibility the industry as a whole could move to Copilot-style pay-as-you-go billing. In that flow, models with high token efficiency could have an advantage in price competitiveness, some observers said. One user shared that they processed 15,000,000 tokens for about 7 cents by linking DeepSeek to a GitHub VS Code environment. Copilot's new pricing system shows that, in the AI coding tools market, per-use pricing is emerging as a key competitive factor alongside performance.

First day with GitHub Copilot token based billing. Productive day, and spent very little tokens. 5.3-codex is a really nice model. Also helps when you do very focused and deliberate changes with AI. 161 credits translates to $1.61 so that's a bit under $50 for full month of work pic.twitter.com/7AQA04xGsX