GitHub Copilot Fusion: 6 Admin Moves Before November 30

GitHub Copilot Fusion routes each prompt across models by effort level. See billing, admin controls and 6 moves before the 30% discount ends Nov 30.

TL;DR: GitHub Copilot Fusion is now generally available. Instead of picking a model, developers pick an effort level and Copilot routes each prompt across one or more models. Each model still bills at its own rate, with a 30% discount through November 30, 2026. Before then, set the model pool, the default effort level and budgets.

Your developers no longer have to choose an AI model, and that moves the decision onto your desk. On October 28, 2026, GitHub made GitHub Copilot Fusion generally available on every Copilot plan in Visual Studio Code, the GitHub Copilot app, GitHub Copilot CLI and JetBrains IDEs. The 30% discount on Fusion model costs runs through November 30, which makes November the cheapest month to pilot it and set your defaults.

GitHub Copilot Fusion is a runtime model router and orchestrator: it chooses the models and the execution pattern for each prompt, runs them, and returns a single response. In GitHub's announcement, AT&T describes a pattern every engineering leader will recognize: developers pick a powerful model for a hard task and forget to switch back for routine work. Fusion removes that friction, and with it the developer as the person who decides what each prompt costs.

Choosing a model was a habit. Choosing a default is a policy.

We made the same argument when frontier models reached Copilot Studio: model choice is now a governance decision. Fusion makes that true for every developer in your enterprise.

What is changing with GitHub Copilot Fusion, and when?

Fusion replaces per-task model choice with effort levels, adds admin controls for its model pool and default, and bills each model it uses at standard rates. Here is each change, its timing and what it means for you.

Swipe horizontally to see when each change lands and what it means.

What is changing with GitHub Copilot Fusion, when it lands and what it means for IT leaders
Change	When	What it means
Copilot Fusion in the model picker	Generally available October 28, 2026	Developers choose Copilot Fusion and an effort level instead of a specific model, on all Copilot plans.
Effort levels: Fusion Standard, Fusion High and Fusion Max	Now	The effort level, not the developer, decides how many models a prompt uses and how capable they are.
Execution patterns: Single, Cascade and Critique	Now, chosen per prompt	Some prompts get a second model as a judge or reviewer, which can raise quality and credit use.
Admin control of the Fusion model pool	Now	Fusion only uses models your policies allow, and you can remove a model from Fusion without removing it for everyone.
Fusion as the default model, with a default effort level	Now, in enterprise managed settings	You can make Fusion Standard the starting point for every new conversation and lock or delegate the effort level.
Billing per model, with a 30% discount on model costs	Discount through November 30, 2026	Paid plans pay less for Fusion in November. From December 1, every model it uses bills at full rate.

How do Copilot Fusion effort levels route and orchestrate models?

The developer chooses an effort level, and Fusion chooses the models and an execution pattern for each prompt. Different prompts in the same session can take different paths.

Fusion Standard balances cost, quality and speed and selects one model for each task. It is built for everyday work.
Fusion High adds model calls when a task is worth the extra time and AI credits, such as a complex bug or a change across several files.
Fusion Max favors the most capable models for the deepest reasoning.
Fusion Fast isn't selectable; it runs only in Visual Studio Code inline chat.

Behind each effort level sit three execution patterns:

Single: one model completes the task, just like a request to one model.
Cascade: an efficient model drafts, and a judge accepts the draft or escalates to a stronger model.
Critique: one model drafts while several others review it in parallel, then the original model revises.

Critique applies a principle we explored in treating AI models as insider risks: no model should be the only check on its own work. Developers can hover over the effort level under a response to see which models ran.

One detail belongs in every developer briefing: if Fusion discards a draft, file edits it already made aren't undone, so review changes before committing. GitHub reports that Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within two points of a leading frontier coding agent, at about half the cost per task. Treat that as vendor data and test it on your own code.

How is GitHub Copilot Fusion billed?

GitHub Copilot Fusion has no charge of its own: you pay for each model it uses, at that model's standard rate, in GitHub AI Credits. Because Fusion High and Fusion Max can use more than one model, a task can cost more than it would on Fusion Standard.

The unit: one AI credit equals $0.01. Inline code completions and next edit suggestions aren't billed in credits.
The pool: Copilot Business includes 1,900 AI credits per user per month and Copilot Enterprise 3,900, pooled across the organization or enterprise. Unused credits don't carry over.
The overflow: additional usage is on by default through the AI credits paid usage policy, billed at published per-token rates.
The discount: through November 30, 2026, paid plans get 30% off model costs when using Fusion.

The default matters most for one group. GitHub Enterprise customers working with a GitHub or Microsoft account team can now get Copilot access included, with no monthly credits, so every request draws on your budget from the first prompt. It is the same lesson Microsoft 365 Copilot leaders are learning: a meter needs a business owner, not just an admin setting.

Why does Copilot Fusion matter for IT leaders?

It turns model choice from a developer habit into an admin policy, and that policy now sets your AI coding bill and your data exposure. Three things change.

Cost: the default effort level becomes your biggest spending lever. A team that starts every conversation on Fusion Standard spends very differently from one that starts on Fusion Max.
Risk: Cascade and Critique can involve several models in one answer, so your data protection approvals must cover every model in the Fusion pool, not just the one a developer used to pick.
Quality: review patterns can catch problems before a developer sees the answer, which is where the hours come from.

If your teams already use agent mode for work like upgrading Java and .NET applications with GitHub Copilot, Fusion High is likely to become their working default. In healthcare and financial services, check the pool against your approved vendor list first.

Nonprofit: small development teams on fixed grants should lock the default at Fusion Standard and set user-level budgets, because additional usage is on unless someone turns it off.
Education: universities can map faculties or research grants to cost centers so heavy research use doesn't drain a shared teaching budget.
Government: Fusion won't use models excluded by data-resident or FedRAMP-compliant model policies, and it doesn't appear in the picker if none of its models are available. Confirm which models remain before you promise developers a default.

Which effort level should each team use by default?

Start everyone on Fusion Standard, give Fusion High to the teams whose work justifies it, and treat Fusion Max as a request, not a default. Use this policy map as your starting point and give every row an owner.

Swipe horizontally to see each team's default, how it is set and its guardrail.

2toLead Copilot Fusion effort level policy map: default effort level, how it is set and the guardrail for each type of team
Team or type of work	Default effort level	How it is set	Guardrail
Most developers, everyday coding	Fusion Standard	Enterprise default, overridable per conversation	User-level budget with alerts
Platform, architecture and modernization teams	Fusion High	Enterprise team override	Cost center budget reviewed monthly
Security fixes, incidents and complex migrations	Fusion Max, on request	Chosen per conversation by named engineers	Time-boxed approval and a written reason
Students, interns and contractors	Fusion Standard	Locked, not overridable	Low user-level budget with a hard stop
Regulated code in healthcare, financial services or government	Fusion Standard or High	Team override plus a restricted Fusion model pool	Only models your data protection review has cleared

What should IT leaders do before November 30?

Make six moves, most of them a few hours each: name owners, set the model pool, set the default, set budgets, brief developers and measure a November pilot while the discount applies.

Name two owners. One for the Fusion model pool, usually platform engineering with security, and one for the AI credit budget, usually the engineering leader with finance. Write both names on the policy record.
Set the Fusion model pool. In your enterprise, select AI controls, then Copilot in the sidebar. Review the baseline under Configure models, then under Models select Configure Copilot Fusion models and disable any model your data protection review hasn't cleared. Check whether the Default availability policy lets unconfigured models in.
Make Fusion Standard the default. In managed-settings.json, set "model": "fusion" and "fusionTier": { "overridable": "standard" }, then give selected enterprise teams "fusionTier": "high" in their own settings file. The setting needs Visual Studio Code 1.144.0, GitHub Copilot CLI 1.0.104-3 or GitHub Copilot app 1.1.33-4 or later.
Set budgets before usage grows. Decide whether the AI credits paid usage policy stays on, add user-level budgets, create a cost center for your pilot group and turn on Stop usage when budget limit is reached for every spending limit.
Brief developers in one page. Explain when to choose Fusion High or Fusion Max, that discarded drafts may leave file edits behind, and that research preview selections have already moved to Fusion.
Measure a November pilot. Under Billing & licensing > AI usage, group usage by model and compare the pilot group with October. Decide by early December whether the default holds at full price.

Microsoft 365 leaders face the same pattern in the Copilot Studio harness, where harness choice is a design decision, not a toggle. Defaults set early tend to stick.

How do you measure the value of Copilot Fusion?

Compare AI credits per developer and per task before and after the default changes, then count the rework hours that review patterns avoid. Here is an illustrative example; replace every figure with your own.

A 150-developer team on Copilot Business shares a pool of 285,000 AI credits a month, worth $2,850. Each developer runs about 60 chat and agent tasks a month, or 9,000 tasks. Today most developers leave a powerful model selected, at an assumed 60 credits a task: 540,000 credits, so 255,000 credits ($2,550) a month is billed above the pool.

With Fusion Standard as the default and full model rates, assume 75% of tasks are everyday work routed to versatile models at 25 credits each, and the other 25% run on Fusion High at 75 credits each because review steps add calls. That totals 337,500 credits, and the overage falls to 52,500 credits ($525). The meter drops by about 38%, and spend above the pool falls by $2,025 a month, about $24,300 a year.

Now the hours. Assume Fusion High's review catches a problem on 1 in 10 hard tasks that would otherwise cost a developer 30 minutes of rework. That is 225 tasks, or about 112 hours a month: roughly $10,100 at a $90 loaded hourly cost.

Track four numbers monthly: AI credits per developer, the share of prompts by effort level, credits per merged pull request and changes reverted after merge. Report them alongside the Copilot adoption metrics that prove value so finance sees one story across Microsoft 365 Copilot and GitHub Copilot.

What do IT leaders ask about GitHub Copilot Fusion?

The short answers: it is a model router, it bills per model, admins can make it the default and it only uses models you allow.

What is GitHub Copilot Fusion?

GitHub Copilot Fusion is a runtime model router and orchestrator in GitHub Copilot. Developers choose an effort level instead of a model, and Fusion picks the models and an execution pattern (Single, Cascade or Critique) for each prompt, then returns one response.

How much does Copilot Fusion cost?

There is no separate charge for Copilot Fusion. You pay for each model it uses at that model's standard rate in GitHub AI Credits, and Fusion High or Fusion Max can use more credits per task. Through November 30, 2026, paid plans get a 30% discount on model costs.

Can administrators set Copilot Fusion as the default model?

Yes. In enterprise managed settings, set model to fusion and fusionTier to standard, high or max. A fusionTier value locks the effort level, or you can make it overridable so enterprise teams set their own. Users can still pick a different model per conversation.

Does Copilot Fusion use models our policies have disabled?

No. Fusion only chooses from models enabled in your enterprise and organization model policies and your plan. Enterprise administrators can also disable a model for Fusion alone under Configure Copilot Fusion models, without disabling it for everyone.

How can 2toLead help?

2toLead helps organizations in healthcare, nonprofit, education, financial services, professional services, government and manufacturing turn AI spend into governed, measurable outcomes. In a fixed-scope, two-week Copilot Fusion Cost Sprint, our Microsoft MVP team will: