GitHub made Copilot Fusion generally available on October 28, the first day of its Universe conference. Fusion is a setting in Copilot's model picker that chooses which AI models handle each prompt, and sometimes runs two or more of them on one task. Every Copilot plan gets its basic Standard level; the High and Max levels, which can combine models, start at the Pro+ plan for $39 a month.
The launch matters because model choice has become a daily chore. In September, 61% of Copilot users who picked a model themselves used more than one model, according to GitHub's Octoverse 2026 report, which also found that developers start more than 40% of pull requests with an agent, up from 17% in March.
How Copilot Fusion works
A developer picks Copilot Fusion instead of a named model, then an effort level. For each prompt, Fusion picks the models and one of three execution patterns, which GitHub's documentation describes this way:
Single: one model completes the task.
Cascade: an efficient model drafts a solution, and a judge decides whether to accept it or escalate to a stronger model.
Critique: one model drafts while several others critique it in parallel, and the first model revises using their combined feedback.
Standard balances cost, quality and latency and selects one model for each task, according to the documentation. High is meant for substantial jobs such as a complex bug or a change across several files. Max aims at "frontier capability while still considering token efficiency."
The documentation also lists limits. Fusion has no context window of its own, so each step is bound by the model running it. If Fusion discards a draft, file edits that draft already made are not undone, and GitHub tells users to review changes before committing them.
Fusion grew out of Project HydraFusion, a research preview GitHub wrote about on September 4, combined with Copilot's existing auto model selection. Fusion runs in Visual Studio Code, the GitHub Copilot app, Copilot CLI and JetBrains IDEs.
Plans and pricing
Fusion has no fee of its own. Users are billed for each model it calls, at that model's standard rate, and the documentation warns that High and Max "may consume more AI credits" than Standard because they can use more than one model. Through November 30, paid Copilot plans get a 30% discount on model costs when using Fusion.
Plan | Price | Fusion levels |
|---|---|---|
Free | $0 | Standard |
Pro | $10/mo | Standard |
Pro+ | $39/mo | All three |
Max | $100/mo | All three |
Business | $19/user/mo | All three |
Enterprise | $39/user/mo | All three |
Source: SaasBuzz from GitHub's Copilot plans page and Fusion documentation, checked October 30, 2026. Copilot Student also gets Standard only.
The individual plans include monthly AI credits worth $15 on Pro, $70 on Pro+ and $200 on Max, drawn down at $0.01 each. By SaasBuzz's arithmetic, the 30% discount stretches those credits about 1.43 times as far on Fusion: Pro+'s $70 covers $100 of model usage at standard rates until the offer ends. For an individual developer, the step from Standard to High and Max is the move from Pro to Pro+, $29 more a month.
The cost claims
GitHub's case rests on one benchmark. In the Artificial Analysis Coding Agent Index, 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, the changelog said. SaasBuzz has not tested Fusion.
Kyle Daigle, GitHub's chief operating officer, put the saving higher in a post on X, citing "up to 77% lower cost" on the same index, and added a personal endorsement:
Post on X by @kdaigle
GitHub Copilot Fusion gives devs frontier intelligence at up to 77% lower cost and up to 40% faster as shown in Artificial Analysis’ coding agent index. It’s replaced choosing models for me over the last two months.
The two figures differ in kind: "up to" is a best case, while the changelog's "about half" is an average per task. The changelog also quotes a customer, AT&T, on the Standard level:
Previously, developers would often choose a more powerful model for a complex task and forget to switch back for everyday activities.Hari Pokhrel, principal software engineer at AT&T, in GitHub's changelog
What it means for teams that buy Copilot
For administrators, Fusion only draws on models enabled in the enterprise and organization model policies; enterprise admins can remove a model from Fusion without disabling it for everyone, and admins can make Fusion the default at any effort level through enterprise managed settings, according to the changelog.
Fusion puts GitHub against Cursor on automatic routing. Cursor's Auto feature has Cost, Balance and Intelligence modes; its documentation says Balance and Intelligence bill at the routed model's list price on Teams and Enterprise plans, and that "Individual plans have Cost mode only." GitHub, by contrast, offers its multi-model levels to individuals from Pro+.
The open question for buyers is predictability. Because Fusion chooses the models, the cost of a task is set by its routing rather than by the developer, and GitHub publishes per-model rates but no typical per-task cost for each Fusion level. Teams that adopt Fusion during the discount will pay full model rates once the offer ends on November 30.
GitHub also opened an experimental preview of local model routing in Fusion the same day: on the Standard level, an experimental option lets Fusion choose between cloud models and models running on the developer's own machine. SaasBuzz's earlier report covers GitHub's plans for more computing capacity.
