GitHub Copilot users see token-based price increase.

Since the announcement in April this year,GitHub The proposed change in billing for Copilot raises a lot of speculation as to how much use-based AI can cost organizations or individuals compared to a fixed monthly rate for subscriptions.

Based on LLM Only one day after the services shifted to token-based billing, software developers and IT departments have reported their findings online - a short version is that, as of 1 June 2026, the cost of using GitHub Copilot in software development and deployment became more expensive.

How has GitHub Copilot's fee scheme changed?

Although the subscription price has not changed (Copilot Pro $10 per month, Pro+39 per month, $19 per month for each user at the commercial level and $39 per month for each user at the enterprise level), the current price refers to the number of points of various AI models available on the GitHub platform per month. For a typical user, the cost of a credit is a penny and, based on the model variant chosen in the reasoning, the score is deducted on the basis of the chip workload spent by artificial intelligence. Thus, the Copilot Enterprise users receive 3,900 credits per month (US$ 39) and the Copilot Business users receive 1,900 credits per month (US$ 19).

Users will consume their points in the form of tokens whose prices vary according to the power and type of model used. For example, use ChatGPT-5.2 The cost per million entry tokens is US$ 1.75 (a token can be considered a word), while the cost per million output tokens is US$ 14 and the price of the cache input (e.g., information held by LLM, providing a continuous context for a series of queries) is US$ 0.175 per million.

When users run out of assigned points, they can choose to buy more. The code completion and “next-editing” proposals in the IDE of the developers are free of charge, but the costs of the code review process are the same as those of other GitHub Copilot activities.

Will users pay more to use Copilot?

It depends to a large extent on the individual user and to a certain extent on the person you ask. Many users of the commentary section of the GitHub Community Discussion Page, which announced these changes as early as April 2026, reported that their points were depleted much faster than expected. The user “rvs99” says, “My 12% AI score is exhausted in very small tasks. I used it as usual. Claude. In response, Sonnet 4.6, it has updated almost only two to three rows of a total of six documents, at a cost of approximately $0.35 per line.” “Prhost” has released screenshots of their account dashboard, showing that there are still 3,705 points left in the 7,000 quota after one day and stating that “it would be easier to close the project. [ Chuckles ]MicrosoftLift up the stones and hit your feet."

User “zoomp05” summed up the tone of most commentators: “The strategy is clear, but it's best to say from the beginning” It's a subsidy test' or something like that to promote our tools.”

The original subscription service provided by GitHub is now out of service and Microsoft, the owner of the platform, may consider it a losing leader. Clearly, it is not sustainable to allow users to burn tokens that are much more valuable than their subscriptions. A rough reading of the announcements and posts of large model providers on the Internet shows that, as a business model, subscription-based billing can only be temporary. It may be surprising that many users are surprised to find that their coding platform is now at a level consistent with the costs of suppliers. Operating LLM is not a cheap undertaking, especially in view of the additional costs involved in developing new models, post-training, maintenance, data centre construction and future loan repayments.

What can a company do now?

Those who invest in using LLM-based coding tools to support their development teams may consider the following options:

- Reassess the return on investment from the artificial smart coding platform and adjust budget allocations accordingly.

- Consider which processes for software development workflows may be cheaper to transfer to artificial intelligence (e.g., junior developers-level code creation) and which are cost pools (code review, multi-agent workflows, fast-paced operations, etc.).

- Finding alternative, less costly platforms. These are divided into three main camps:

- Open model of local hosting. These are not front lines.LLMThere is also a lack of many of the coding “tools” provided by the professional coding platform.

- Hosting of a cutting-edge model from a Chinese provider of law and a Master of Laws, such as Ali Baba.

- “Auxiliary” coding platforms such as Cursor may provide temporary breathing opportunities, but note that many alternatives come from OpenAI and Anthropic The front-line, more well-known model is likely to be the same as GitHub Copilot.

(Pornography source: Pixabay, licensed.

Would you like to learn more about artificial intelligence and big data from industry leaders? View the AI Fair in Amsterdam, California and London. The event was part of TechEx and was held in conjunction with other leading technical events. Click here for more information.

AI News is supported by TechForge Media. Other upcoming business technology activities and networking workshops are explored here.