GitHub Copilot Fusion changes the most annoying decision in AI coding: which model to use. On October 28, 2026, at GitHub Universe, GitHub made Fusion generally available on every Copilot plan. You pick an effort level — Standard, High, or Max — and Fusion picks the models and decides how they work together.
For freelancers who bill by the project, that is a cost story first. GitHub says Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within two points of Claude Code with Opus 5.5 Max, at roughly half the API cost, half the tokens, and 40% less time.
What Fusion actually does
When you select Copilot Fusion in the model picker, you are not choosing a model. GitHub’s changelog calls it “a system that routes and orchestrates models for each task.” Three patterns do the work:
- Single. One model finishes the job when that is enough.
- Cascade. An efficient model drafts, a quality check accepts it or escalates to a stronger model.
- Critique. One model drafts, another reviews, and the first revises.
The three effort levels set the ambition. Fusion Standard prioritizes versatile models for everyday coding. Fusion High prioritizes capable models for harder tasks. Fusion Max prioritizes the most capable models available to your plan.
Fusion works in Visual Studio Code, the GitHub Copilot app, Copilot CLI, and JetBrains IDEs today. Developers who enabled the auto, HydraFusion, or HydraFusion Max previews move automatically to the matching Fusion level.
The evidence: score, billing, and the discount deadline
| Detail | What the source says |
|---|---|
| Announced | October 28, 2026, at GitHub Universe 2026, San Francisco |
| Availability | Generally available on all Copilot plans |
| Clients | VS Code, GitHub Copilot app, Copilot CLI, JetBrains IDEs |
| Effort levels | Standard (versatile models), High (capable models), Max (most capable models) |
| Benchmark claim | Fusion Max scored 60 on the Artificial Analysis Coding Agent Index (GitHub’s own test) |
| Comparison claim | Within 2 points of Claude Code with Opus 5.5 Max, at about half the API cost, half the tokens, 40% less time |
| Billing | Billed per model at each model’s standard rate; no separate Fusion fee |
| Discount | 30% off model costs for paid Copilot plans, limited time; secondary coverage of GitHub’s docs names November 30, 2026 |
| Admin controls | Admins can restrict which models Fusion may use, exclude models, and set Fusion as the default via enterprise managed settings |
A coding agent index benchmarks AI coding tools on real tasks. Model routing means the system, not you, picks the model per prompt, trading a little control for lower cost.
How Fusion handles one coding task
1. You pick effort — Standard, High, or Max in the model picker.
2. Fusion drafts — A single model, or an efficient model first in a cascade.
3. Quality check or critique — A second model reviews, or the task escalates to a stronger model.
4. You review the result — Especially any file edits from a discarded draft, which are not auto-undone.
5. You pay per model used — At standard rates, minus 30% on paid plans during the discount window.
How to profit from it
Stop paying premium-model prices for routine code. Here is the playbook:
- Run the comparison this week. Take one real task you would send to a premium model, run it on Fusion Standard and on your usual model, and compare the result and your usage report. If the cheaper path holds, make it your default.
- Use Max only for hard tasks. Keep the most capable tier for architecture, debugging, and refactors, and let Standard chew through boilerplate, tests, and docs.
- Bank the 30% discount. The discount window is reported to end November 30, 2026. Do your comparison now, not in December, or you pay full price while you decide.
- Sell faster delivery, not cheaper hours. If Fusion cuts time per task near GitHub’s claim, quote shorter timelines at the same price and keep the margin.
What to watch before you switch
First, the benchmark is GitHub’s own test, and it covers only the Max tier against one named competitor. Standard and High have no published comparison, so verify them on your own tasks.
Second, High and Max can call more than one model per task, so one request can cost more credits than Standard. Watch your usage report in week one.
Finally, Fusion works separately from subagents and does not undo edits from discarded drafts. Review the workspace before every commit, the same way you would with any agent.
Sources: GitHub’s October 28 changelog as reported by Subagentic and FourWeekMBA, linked below. Benchmark and cost figures are GitHub’s own claims and were not independently verified.
Frequently Asked Questions
What is GitHub Copilot Fusion?
A generally available system in Copilot that routes and orchestrates models for each coding task. You pick an effort level — Standard, High, or Max — and Fusion picks the models.
How is Fusion billed?
Per model, at each model’s standard rate, with no separate Fusion fee. Paid Copilot plans get a 30% discount on those model costs for a limited time.
When does the 30% discount end?
GitHub calls it limited-time; secondary coverage of its docs names November 30, 2026, though the changelog itself named no end date.
How good is Fusion, really?
GitHub says Fusion Max scored 60 on the Coding Agent Index, within two points of Claude Code with Opus 5.5 Max, at about half the API cost. That is GitHub’s own test.
Where does Fusion work?
Visual Studio Code, the GitHub Copilot app, Copilot CLI, and JetBrains IDEs, on all Copilot plans, with more clients coming.
Sources: Subagentic — Copilot Fusion is generally available · FourWeekMBA — Copilot Fusion nears Opus 5.5 at half the cost
Keep Reading
Want more tools that actually pay for themselves? Browse the AI Tools section for the latest launches, pricing breakdowns, and profit playbooks — and check the official source linked above before you buy or deploy anything.