GitHub Copilot Fusion is a new router inside Copilot that decides which AI model, or mix of models, should handle each task, so you don’t have to pick one yourself. GitHub announced it at Universe 2026 on October 28, alongside a new Copilot app and a billing change for Enterprise customers.
If you’ve ever wondered whether a bug needs the “smart” model or the cheap one, this is GitHub’s answer: let the system choose.
Here are the quick facts before we get into the details.
- Announced: October 28, 2026, at GitHub Universe
- What it does: routes each task to one or more models based on what the task needs
- GitHub’s claim: Fusion Max lands within two points of Opus 5.5 Max at half the cost and 40% less time
- Billing: Enterprise customers can use all Copilot features and pay only for AI credits used
- Not yet public: per-credit prices and Fusion plan details
Treat the benchmark as GitHub’s own number for now. Nobody outside the company has re-run it yet.
What GitHub Copilot Fusion actually does
Fusion looks at a request and judges what it needs: reasoning, code generation, debugging, or tool use. Then it picks how to run the job, on one model or several.
That matters because the old workflow was manual. You chose a model from a dropdown, guessed wrong, and either overpaid or got a weak answer.
According to the GitHub Universe 2026 announcement, the top tier, Fusion Max, scored within two points of Claude Opus 5.5 Max on the Artificial Analysis Coding Agent Index. GitHub says it did so at half the cost and with 40% less time per task.
The new Copilot app and Copilot projects
The bigger change for daily work is the new GitHub Copilot app. It connects agents and tools across projects in one place, and it isn’t limited to code. GitHub says it also covers writing specs, design discussions, and analyzing telemetry.
Inside it, Copilot projects went live on October 28. They let agents work across several repositories in parallel, managed sessions, with one place to track progress and make decisions.
This lines up with a trend we’ve covered before. If you followed our look at Cursor’s computer-use agents, you’ll see the same idea: the editor is turning into a control room for many agents.
Copilot automations and code review Max
Copilot automations are cloud-run workflows tied to a repo. You can start them by hand, trigger them from an event, or put them on a schedule. Typical jobs are issue triage and code review.
The new part is that automations can now learn from how past runs went and propose improvements. A human still has to approve those changes.
Code review also gets a third level. Lite and Balanced handle everyday reviews, and the new Max tier is for changes that need a deeper look. GitHub says Max posted the highest F1 score on ReviewBench for high-severity issues.
Agents that run on schedules also raise safety questions. Our piece on the AGENTS.md attack is worth a read before you give any agent broad repo access.
Who pays what: the Enterprise credits change
GitHub Enterprise customers can now use every Copilot feature, including the app, code review, and automations, without a separate Copilot subscription. They pay for the AI credits they consume instead.
It’s available now through GitHub Sales, and self-serve is listed as “coming soon.” Existing Copilot Business and Copilot Enterprise customers can stay on their current plans or switch later.
Here’s how the pieces compare at a glance.
| Piece | What it is | Status |
|---|---|---|
| Copilot Fusion | Routes tasks across one or more models | Announced; plan details not stated |
| Copilot app | One place for agents across projects | Announced; pricing not stated |
| Copilot projects | Parallel agent sessions across repos | Live from October 28 |
| Code review Max | Deeper tier next to Lite and Balanced | Announced; pricing not stated |
| Enterprise AI credits | Pay per credit, no separate seat | Via GitHub Sales now; self-serve soon |
Why does that matter? Usage-based billing rewards efficient routing and punishes sloppy prompts. If Fusion really sends easy jobs to cheaper models, your credit burn drops. If it over-picks the big model, it climbs fast, and you won’t see that until the invoice arrives.
The gaps are the story here. GitHub hasn’t published a per-credit price, so you can’t yet estimate what a month of agent work will cost.
Frequently Asked Questions
What is GitHub Copilot Fusion?
It’s a routing layer in Copilot that assesses a task and runs it on one or more models. You don’t choose the model yourself.
Is Copilot Fusion free?
GitHub hasn’t stated plans or pricing for Fusion in its announcement. Enterprise customers pay through AI credits for what they use.
Do I need a separate subscription on GitHub Enterprise?
No. Enterprise customers can use all Copilot features and pay only for the credits they consume. Self-serve sign-up is still “coming soon.”
Can I trust the Fusion benchmark?
It’s GitHub’s own result on the Artificial Analysis Coding Agent Index. Wait for independent tests before you rely on the cost and speed claims.
Our take: worth a trial, not a switch
Model routing is a sensible idea, because most developers don’t want to babysit a model picker. But without credit prices, Fusion is a promise rather than a plan. Try it on a few real tickets, track what credits it burns, and compare that to your current setup before moving a team over.
Leave a Reply