Playbooks · Coding tasks
Copilot Fusion is generally available: what it does to your AI-credit bill
Written October 31, 2026
For a Copilot user turning on Fusion: you stop picking a model and pick an effort level instead, and Copilot routes each task to one or more models. That routing can save credits on everyday work and spend more of them on hard work — the bill follows how many models answer, not the effort level you picked.
What changed
GitHub made Copilot Fusion generally available on 28 October 2026. Instead of selecting a model in the model picker, you select Copilot Fusion and choose an effort level: Fusion Standard, Fusion High, or Fusion Max. Fusion then chooses the execution pattern and the models for each prompt. It is available in Visual Studio Code, the GitHub Copilot app, GitHub Copilot CLI, and JetBrains IDEs, on all Copilot plans, according to GitHub's changelog. Fusion brings together Project HydraFusion's multi-model orchestration and Copilot auto model selection's model selection, per the same changelog.
Fusion chooses one of three execution patterns per prompt, as GitHub's Fusion documentation describes them: Single, where one model completes the task; Cascade, where an efficient model drafts a solution and a judge decides whether to accept it or escalate to a stronger model; and Critique, where one model drafts a response while multiple models critique it in parallel, and the original model revises from the combined feedback. Different prompts in one session can use different patterns. Fusion only uses models enabled in your plan and your organization's policies, and administrators can set Fusion as the default for users under Enterprise managed settings.
How the billing works
There is no separate charge for Fusion itself. GitHub's documentation states the rule plainly: when you use Copilot Fusion, you are billed for each model it uses, at that model's standard rate. Because Fusion High and Fusion Max can use more than one model for a task, a task may consume more AI credits than it would with Fusion Standard. Fusion Standard selects one model for each task, so a straightforward task under Standard behaves like a request to one model. The documentation also states that Fusion only adds model calls when a task warrants them, so a straightforward task does not wait or use credits for a review it does not need, and that routing happens along natural cache boundaries to avoid additional cache-related costs.
AI credits are the billing unit: 1 AI credit equals $0.01 USD, and each interaction is priced by the tokens it consumes at the rate of the model used. Monthly allowances on individual plans, from GitHub's plans page, are 1,500 credits for Copilot Pro ($10 a month: 1,000 base credits plus a 500-credit flex allotment), 7,000 for Copilot Pro+ ($39), and 20,000 for Copilot Max ($100). Unused included credits do not carry over between months. There is no published fixed credit charge for a Fusion task, because the charge depends on which models Fusion calls and how many tokens each one uses.
Through 30 November 2026, users on paid Copilot plans receive a 30% discount on model costs when using Copilot Fusion, according to the Fusion documentation. The changelog calls the discount limited-time without naming an end date; the documentation names 30 November 2026. Treat the discount as a window that ends on that date, not as a standing price: after it ends, the same Fusion task costs its full per-model rates.
Strengths
- No separate Fusion fee. You pay the standard rate of each model Fusion uses, and nothing extra for the routing itself.
- Standard stays single-model. Fusion Standard selects one model for each task and balances cost, quality, and latency, so everyday work does not pay for a review it does not need.
- GitHub's cited benchmark result. In the Artificial Analysis Coding Agent Index, GitHub's changelog states that Fusion Max scored 60, within two points of Claude Code with Opus 5.5 Max, at about half the average API cost per task, using about half as many tokens and taking about 40% less time per task. That is GitHub's citation of the index, not an independent measurement, and it describes Fusion Max against one rival setup — it does not promise the same saving on your tasks.
- One quoted efficiency gain. GitHub's changelog quotes Hari Pokhrel, a principal software engineer at AT&T, saying Fusion Standard improved token efficiency there by selecting the right model for each task, so developers were less likely to stay on a more powerful model after a hard task. That is one organization's reported experience, quoted by GitHub, not a measured result for Copilot users generally.
Weaknesses and limits
- High and Max can cost more than Standard on the same task. A Cascade or Critique run calls more than one model, and every model call is billed at its standard rate. The extra calls buy a draft, a review, or an escalation; they are not free, and the discount window only softens them until 30 November 2026.
- No per-task price to budget against. Fusion does not publish a credit price per task, because the pattern and the models are chosen per prompt. Two similar-looking prompts can take different patterns and cost different amounts.
- Discarded drafts are not undone. The documentation warns that if Fusion discards a draft, changes that draft already made in your workspace, such as file edits, are not undone automatically. Review changes before committing them.
- The model pool changes. The documentation states that the model pool changes as new models become available and as GitHub evaluates which models perform best, and that execution patterns may change over time. A task's cost under Fusion today is not a fixed property of the task.
- Credit pools are monthly and do not carry over. A Pro allowance of 1,500 credits is $15 of model usage at published rates. A run of Fusion Max tasks that each call several models can use a visible share of that allowance in one session, and extra usage beyond the allowance is billed unless an administrator has disabled paid usage.
What to do about it
- Start on Fusion Standard. It selects one model per task, so its cost behaves like picking one model yourself, and it is the level where the routing saves credits by moving you off a powerful model after a hard task.
- Check your usage dashboard after your first High or Max tasks. GitHub's billing documentation states that the usage dashboard shows your available allowance and what you have used. Compare the credits a task consumed against the same kind of task under Standard before making High or Max your default.
- Do not budget on the 30% discount past 30 November 2026. Price any ongoing Fusion use at full per-model rates, and recheck the documentation after that date rather than carrying the discount forward.
- Review every Fusion session's changes before committing. A discarded draft's edits stay in your workspace. Read the diff before the commit, whatever effort level ran.
- Keep the effort level per task, not per week. Use Standard for everyday work, and reserve High or Max for a substantial, well-scoped task where extra model calls may be worth the extra time and credits, as the documentation frames those levels.
Limits of this evaluation
- This page is grounded in GitHub's changelog and documentation. No Copilot task was run under Fusion for this page, so no measured saving or measured credit cost for a real task is claimed here. The benchmark and efficiency figures are GitHub's own statements, carried with that attribution.
- Model retirements and per-model rates change over time. The per-model rates Fusion bills at are on GitHub's models and pricing page, not fixed in this page, and the model pool Fusion draws on changes as GitHub evaluates models.
Related playbooks
- Gemini Computer Use is generally available
- Decide whether Copilot's assisted permissions should approve anything for you
- Copilot CLI 1.0.106: /code, /settings, GPT-6.1 Terra
- Copilot CLI 1.0.108: index, PDFs, Ctrl+C
Sources
- https://github.blog/changelog/2026-10-28-route-and-orchestrate-models-with-github-copilot-fusion/
- https://docs.github.com/en/copilot/concepts/models/copilot-fusion
- https://docs.github.com/en/copilot/about-github-copilot/subscription-plans-for-github-copilot
Back to: Coding playbooks hub · Also: Review a pull request with a coding agent · Onboard onto an unfamiliar codebase with an agent · Agent cost estimator