AI costs how much? GitHub Copilot users react to new usage-based pricing system.
GitHub Copilot’s new credit-based billing has triggered sticker shock as users compare ordinary work to high monthly costs.
Intelligence analysis by GPT-5.4 Mini
GitHub has replaced Copilot’s request-based billing with usage-based AI credits, and many users say the change makes normal work far more expensive than they expected. The article highlights both the backlash and the behavior changes some users are already adopting to control costs.
GitHub changed how Copilot charges for AI help. Before, many people got a set amount of usage. Now, the company counts how much AI work each task uses, like counting pennies for every piece of help.
That surprised a lot of users. Some found that normal coding tasks used up their monthly limit much faster than expected, kind of like a game where each move suddenly starts costing coins.
A few people are still making it work by using smaller requests and cheaper AI models. But others worry the tool may become too expensive for heavy use.
Analysis
What changed
GitHub’s new Copilot pricing model moves paid subscribers from request-based allowances to monthly AI credits. Each credit equals one cent of usage, and the amount spent depends on the token cost of the model and the size of the request.
The article says the old system treated very different workloads the same, whether someone asked a quick question or ran a long autonomous coding session. GitHub argued that this setup pushed much of the inference cost onto Copilot itself. Under the new plans, Pro, Pro+, and Copilot Max now come with fixed monthly credit bundles that translate to $15, $70, and $200 worth of usage respectively.
User reaction
The immediate reaction has been a wave of surprise and frustration. Users on social media and forums shared screenshots and estimates showing that a day of ordinary use can burn through a large share of a monthly allowance, and in some cases the reported usage would have cost thousands of dollars under the new model.
Ars notes that the cost depends heavily on the chosen model. Cheaper models can be relatively modest, but frontier models can multiply the bill quickly, especially when Copilot’s “Auto” mode selects a more expensive model than the user expected. The article also points out that long chat sessions can quietly get pricier because the whole conversation history becomes part of the input.
The broader shift
Not everyone sees the change as a disaster. Some users say they are adapting by keeping prompts narrower, choosing cheaper models, and avoiding long-running chats. But the article suggests the larger industry may move this way too, replacing subsidized AI access with more explicit usage pricing.
If that happens, token efficiency becomes a competitive advantage. The story closes by pointing to users already looking at cheaper setups outside Copilot, including self-managed integrations with other models.
Key points
- GitHub Copilot has switched from request-based billing to monthly AI credits.
- Users report that ordinary usage can burn through credits far faster than expected.
- Copilot costs now vary by model choice, token count, and chat history length.
- Some users are already adjusting their habits to reduce usage and cost.
- The article suggests usage-based pricing may spread across the AI coding market.
If the new system nudges people toward more careful prompt use, Copilot could become more predictable and more closely matched to actual consumption. The article also suggests that users who learn to choose efficient models and tighter workflows may still find the service useful at a manageable cost.
The biggest risk is that users who relied on heavy Copilot usage may feel priced out and leave for other tools. The article also shows that model choice and long chat histories can make costs jump quickly, which could create distrust and push developers to avoid the most capable models.
