GitHub to switch Copilot to usage-based AI billing in 2026

GitHub will move Copilot from request-based billing to usage-based billing on June 1, 2026, citing unsustainable inference costs. The company will keep subscription prices the same but add monthly GitHub AI Credits tied to token consumption.

GitHub to switch Copilot to usage-based AI billing in 2026

Microsoft-owned GitHub is changing how it bills GitHub Copilot, moving from request-based pricing to usage-based billing on June 1, 2026. The company says the current model, which lets subscribers spend a fixed number of premium requests, no longer matches the cost of serving more complex AI workloads.

In a blog post, Mario Rodriguez, GitHub’s chief product officer, said the company has been absorbing much of the rising inference cost behind Copilot usage. Inference cost is the compute expense of running an AI model to produce an answer, and GitHub says the existing premium request model is no longer sustainable.

⚡ New to this?

This news is about how a major AI coding product is changing from a flat-request plan to a metered one. Tokens are the small pieces of text AI systems process, and usage-based billing means customers pay more directly for how much the model actually works.

For non-experts, the key issue is that AI tools are expensive to run, especially when they act like coding assistants or autonomous agents that work for long periods. GitHub is signaling that it can no longer absorb those costs inside subscription fees.

🦞 OpenClaw angle

If you run self-hosted agents or build automation on top of AI APIs, start tracking token use per task now, not just per request. Put hard budgets and per-job caps around long-running agent loops, and simulate worst-case prompts before you commit to a pricing model.

You should also separate lightweight autocomplete-style features from heavy reasoning workflows, because GitHub’s move shows those two usage patterns do not cost the same. If your system exposes metering to customers, add a preview billing screen or usage estimator before you roll out agents that can run for hours.

Under the current system, a quick chat question and a long autonomous coding session can count the same against a user’s plan, even though they cost GitHub very different amounts to serve. Rodriguez said that creates a mismatch between what customers pay and what the company spends.

The new model is based on token usage. Tokens are small chunks of text, often only a few characters, that AI models use to measure input and output. GitHub will convert that activity into a virtual currency called GitHub AI Credits, with one credit equal to $0.01.

According to GitHub, Copilot plans will include a monthly allotment of AI Credits, and paid users will be able to buy more usage if they run out. The company said usage will be calculated from input, output, and cached tokens, using the listed API rates for each model. Because different models and prompts consume tokens differently, GitHub said the final bill can vary in ways that are harder to predict than the old request system.

To help customers prepare, GitHub plans to introduce a preview bill experience in early May. That tool will give users and administrators visibility into projected costs before the June 1 transition.

The change comes as AI providers face growing pressure from heavy use of coding tools and autonomous agents. The article says AI companies were surprised by the attention OpenClaw attracted in February, which pushed more experimentation with agents that run continuously on tasks. It also says improved model capability encouraged more developers to use AI coding tools, increasing demand beyond what some subscription plans were designed to cover.

GitHub recently signaled that pressure by suspending new Copilot, Pro, Pro+, and Student plans. The article says Anthropic and Google have also limited some uses of their services, while OpenAI introduced a more expensive $100 tier for its Codex model and is considering an end to unlimited usage under subscription plans.

GitHub’s subscription prices will stay the same. Copilot Pro remains $10 per month, Pro+ stays $39 per month, Business remains $19 per user per month, and Enterprise remains $39 per user per month.

At the $0.01 AI Credit rate, Copilot Pro subscribers get 1,000 credits a month and Pro+ subscribers get 3,900. Organizations and enterprises receive 1,900 and 3,900 API Credits per user per month, respectively, though existing Business and Enterprise customers get higher allotments from June 1 through September 1, 2026, at 3,000 and 7,000 respectively.

Users on annual plans can cancel for a prorated refund or be downgraded to Copilot Free when the plan expires, and those annual plans will not be renewable. The article says premium model multipliers will also rise under the new regime: Anthropic’s Opus 4.7 will move from 7.5x to 27x, while OpenAI’s GPT-5.4 will rise from 1x to 6x.

GitHub said code completions and Next Edit Suggestions will remain unlimited on paid plans even after users hit their AI Credit limit. Other Copilot features will be governed by the new metered system when the change takes effect on June 1, 2026.

Source: RedPacket Security ↗

More from AI News