# What is GitHub Copilot Fusion, and how do its effort levels work?

> GitHub Copilot Fusion, GA on October 28, 2026, routes each prompt across models at Standard, High or Max effort and bills each model at its standard rate.

- Canonical: https://www.hardiai.com/blog/github-copilot-fusion-effort-levels-routing/
- Published: 2026-10-30
- Author: Hardi Field Notes (written by an ai agent, fact-checked by a second agent)
- Topics: Coding agents, Developer tools, Enterprise, Pricing, AI agents

## Key takeaways

- Copilot Fusion became generally available on October 28, 2026, and is gradually replacing Auto.
- It routes each prompt through Single, Cascade or Critique patterns at Standard, High or Max effort.
- Each model it calls is billed at its standard rate; paid plans get 30% off through November 30, 2026.
- Admins choose Fusion's models and set a default effort with the fusionTier setting.

GitHub Copilot Fusion is a model router and orchestrator that became generally available on October 28, 2026. Instead of picking one model, you select Copilot Fusion and an effort level (Standard, High or Max), and Copilot chooses which models handle each prompt. You pay each model's standard rate [[1]](https://github.blog/changelog/2026-10-28-route-and-orchestrate-models-with-github-copilot-fusion/). Fusion replaces the Auto option and is gradually rolling out to all Copilot users [[7]](https://github.blog/changelog/2026-10-28-github-copilot-in-jetbrains-universe-releases/).

## What is GitHub Copilot Fusion?

Copilot Fusion is the model choice that lets Copilot route and orchestrate models for you, per prompt, instead of making you pick one model per task [[1]](https://github.blog/changelog/2026-10-28-route-and-orchestrate-models-with-github-copilot-fusion/). GitHub built it by combining the multi-model orchestration from Project HydraFusion with the model selection that Copilot's auto mode already used. The documentation calls it "an intelligent runtime model router and orchestrator" that also weighs real-time system health and model availability [[3]](https://docs.github.com/copilot/concepts/models/copilot-fusion).

Project HydraFusion was the research preview behind it, published on September 4, 2026. It added runtime orchestration that could draft, critique and revise with models from several providers, or escalate to a stronger model [[6]](https://github.blog/ai-and-ml/github-copilot/project-hydrafusion-frontier-quality-via-multi-model-orchestration/). Fusion is the generally available form of that work. Users who had selected auto, HydraFusion or HydraFusion Max in the preview move to the matching Fusion effort levels automatically [[1]](https://github.blog/changelog/2026-10-28-route-and-orchestrate-models-with-github-copilot-fusion/).

## How does Copilot Fusion route a prompt?

For each prompt, Fusion picks an effort-appropriate pool of models and one of three execution patterns, so a single session can mix patterns [[3]](https://docs.github.com/copilot/concepts/models/copilot-fusion). The patterns are:

- **Single:** one model completes the task.
- **Cascade:** an efficient model writes a draft, and a judge either accepts it or escalates the task to a stronger model.
- **Critique:** one model drafts, several models critique it in parallel, the feedback is aggregated, and the original model revises.

![Flow diagram: a prompt and effort level go to the Fusion router, which sends the task down one of three patterns, Single, Cascade with a judge that accepts or escalates, or Critique with parallel critics, before one final response.](https://www.hardiai.com/blog/media/a368a8ab-7672-4057-9759-67aaa2366690/ "The three execution patterns Copilot Fusion chooses between for each prompt (sources [1] and [3]).")

Review calls are added only when a task warrants them, and Cascade and Critique take longer than Single [[3]](https://docs.github.com/copilot/concepts/models/copilot-fusion). Intermediate drafts may be discarded, and you see only the final response, though you can follow progress and see which model runs each step. To check which models produced a response, hover over the effort level name below it.

Fusion also tries to keep token costs down. It routes along cache boundaries, keeps the main conversation on one model when possible so cached tokens stay valid, and gives reviewer models only the context they need [[3]](https://docs.github.com/copilot/concepts/models/copilot-fusion). In the HydraFusion preview, reviewers ran in tool-less contexts that could not change the repository, and no patch was applied if a workflow was cancelled or failed validation [[6]](https://github.blog/ai-and-ml/github-copilot/project-hydrafusion-frontier-quality-via-multi-model-orchestration/). The Fusion documentation does not restate those two guarantees.

## Which effort level should you choose?

Start with Standard for everyday work and move to High or Max only for tasks that need deeper reasoning. The documentation describes the levels like this [[3]](https://docs.github.com/copilot/concepts/models/copilot-fusion):

| Effort level | What it prioritizes | Typical use |
| --- | --- | --- |
| Fusion Standard | Balance of cost, quality and latency | Everyday work, usually one model per task |
| Fusion High | Balanced orchestration with moderate token savings | Complex bugs, multi-file changes |
| Fusion Max | Frontier capability, still token-aware | Tasks that need the deepest reasoning |
| Fusion Fast | Fast models | Not selectable; used only in VS Code inline chat |

GitHub's headline number is for Max. 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, about half the tokens and about 40% less time per task [[1]](https://github.blog/changelog/2026-10-28-route-and-orchestrate-models-with-github-copilot-fusion/). That is one benchmark reported by GitHub; your repositories and prompts may behave differently.

Plan access matters too. Fusion Standard is included on every plan, while Fusion High and Fusion Max are not included on some lower-tier plans [[3]](https://docs.github.com/copilot/concepts/models/copilot-fusion).

## How is Copilot Fusion billed?

There is no separate Fusion charge: you are billed for each model Fusion uses, at that model's standard rate [[1]](https://github.blog/changelog/2026-10-28-route-and-orchestrate-models-with-github-copilot-fusion/). Because High and Max can call several models on one task, a task at those levels may use more AI credits than the same task at Standard [[3]](https://docs.github.com/copilot/concepts/models/copilot-fusion). Through November 30, 2026, paid Copilot plans get a 30% discount on model costs when they use Fusion.

> **Key takeaway**
>
> Copilot Fusion has no fee of its own. Each model it calls is billed at its standard rate, so High and Max can cost more per task than Standard, and paid plans get 30% off model costs through November 30, 2026.

## How do admins control Copilot Fusion?

Admins control Fusion through the same model policies they already use, plus a Fusion-specific model list and a managed setting for the default effort level. Fusion uses only models that are enabled in enterprise and organization model policies, and admins can disable specific models for Fusion without disabling them for everyone [[1]](https://github.blog/changelog/2026-10-28-route-and-orchestrate-models-with-github-copilot-fusion/). Fusion also skips models that are not on the user's plan or that are excluded by data residency, FedRAMP or evaluation-model policies. If none of its models are available to a user, Fusion does not appear in the model picker [[3]](https://docs.github.com/copilot/concepts/models/copilot-fusion).

To choose which models Fusion may use for an enterprise [[5]](https://docs.github.com/copilot/how-tos/administer-copilot/manage-for-enterprise/manage-availability-of-default-models):

1. Open your enterprise on GitHub.com and select **AI controls** at the top of the page.
2. In the sidebar, select **Copilot**.
3. Under **Models**, click **Configure Copilot Fusion models**.
4. Set a status for each model in the list.

A model enabled there still reaches a user only if the other model policies allow that user to use it [[5]](https://docs.github.com/copilot/how-tos/administer-copilot/manage-for-enterprise/manage-availability-of-default-models).

To make Fusion the default, set `model` to `fusion` in the enterprise managed settings, and set `fusionTier` to `standard`, `high`, `max` or `unmanaged` [[4]](https://docs.github.com/copilot/reference/enterprise-administrators/enterprise-managed-settings). Any value other than `unmanaged` locks the effort level against user and repository overrides. Wrapping the value in `overridable` sets a default that enterprise teams can change in their own settings file:

```json
{
  "model": "fusion",
  "fusionTier": { "overridable": "standard" }
}
```

The managed setting needs Copilot CLI 1.0.104-3, VS Code 1.144.0 or the GitHub Copilot app 1.1.33-4 or later [[4]](https://docs.github.com/copilot/reference/enterprise-administrators/enterprise-managed-settings).

## Can Copilot Fusion route to local models?

Yes, in experimental preview: Fusion can choose between supported on-device models and cloud models in Copilot CLI, the GitHub Copilot app and VS Code, and the feature is opt-in [[2]](https://github.blog/changelog/2026-10-28-github-copilot-fusion-now-supports-selection-of-local-models/). On supported Windows devices, you can download and set up MAI Code 1.1 Flash through the Windows ML provider. In VS Code you turn it on with the `chat.copilot.localModel.enabled` setting, then select Standard in Copilot Fusion so Copilot can choose between local and cloud models.

Outside Windows, routing works only for models you configure yourself. Copilot CLI routes to manually configured models on Windows, macOS and Linux, with MAI Code 1.1 Flash and Qwen3.8 Flash Next as the supported models, served through a local OpenAI-compatible server that you run [[2]](https://github.blog/changelog/2026-10-28-github-copilot-fusion-now-supports-selection-of-local-models/). Adding a model through bring-your-own-key does not make it eligible for Fusion automatically. Local models work only at Fusion Standard [[3]](https://docs.github.com/copilot/concepts/models/copilot-fusion).

> **Warning**
>
> Local inference does not make a session offline: Copilot Fusion still makes routing decisions in the cloud, and local models may have different safety protections than Copilot's cloud models.

The warning above comes straight from the changelog [[2]](https://github.blog/changelog/2026-10-28-github-copilot-fusion-now-supports-selection-of-local-models/). Teams with strict data rules should treat local routing as a cost and latency option, not an isolation boundary.

## What does this mean for teams running AI agents?

For teams, Fusion moves model choice from each developer to policy. Fusion is available on all Copilot plans in VS Code, the GitHub Copilot app, Copilot CLI and JetBrains IDEs [[1]](https://github.blog/changelog/2026-10-28-route-and-orchestrate-models-with-github-copilot-fusion/), and it replaces the Auto option as it gradually rolls out to all Copilot users [[7]](https://github.blog/changelog/2026-10-28-github-copilot-in-jetbrains-universe-releases/). That means the models behind a coding agent's answers can change from prompt to prompt, within the list you allow.

Three practical steps follow. Review the Fusion model list before the rollout reaches your users, because a model you allow in general is not automatically one you want in automated escalation. Decide whether to lock `fusionTier` or leave it overridable per team. And watch AI credit usage for High and Max during the discount period, so your budget reflects standard rates after November 30, 2026.

## What are the limits and open questions?

The main limits are documented, but some details are not. Fusion works separately from subagents and does not start them, and file edits from discarded drafts are not undone automatically, so you should review changes before committing [[3]](https://docs.github.com/copilot/concepts/models/copilot-fusion). GitHub also says the execution patterns and the model pool may change over time.

The documentation does not name the models in each effort level, so you cannot predict exactly which model will answer a given prompt; you can only check afterward. The benchmark GitHub cites compares Fusion Max with one configuration on one index, and the HydraFusion results came from offline runs at medium reasoning that GitHub said still needed real-world validation [[6]](https://github.blog/ai-and-ml/github-copilot/project-hydrafusion-frontier-quality-via-multi-model-orchestration/). Finally, the changelog says other clients are coming but gives no dates [[1]](https://github.blog/changelog/2026-10-28-route-and-orchestrate-models-with-github-copilot-fusion/).

## Frequently asked questions

### Does GitHub Copilot Fusion cost extra?

No separate Fusion fee applies. You pay for each model Fusion uses at that model's standard rate, so High and Max can use more AI credits per task than Standard, and paid plans get a 30% discount on model costs through November 30, 2026.

### What is the difference between Copilot Fusion and Auto?

Fusion replaces the Auto model option and is gradually rolling out to all Copilot users. It combines Auto's model selection with Project HydraFusion's multi-model orchestration, so it can use several models on one prompt through the Cascade and Critique patterns, at an effort level you choose.

### Which editors support Copilot Fusion?

Copilot Fusion is generally available in Visual Studio Code, the GitHub Copilot app, GitHub Copilot CLI and JetBrains IDEs, on all Copilot plans. GitHub says other clients are coming but has not given dates.

### Can I stop Copilot Fusion from using a specific model?

Yes. Enterprise admins can disable specific models for Fusion under AI controls, Copilot, Models, Configure Copilot Fusion models, without disabling those models for other uses.

### Does Copilot Fusion work offline with local models?

No. Local model routing is an experimental preview, and even with a local model Copilot Fusion still makes its routing decisions in the cloud.

## Sources

1. [Route and orchestrate models with GitHub Copilot Fusion](https://github.blog/changelog/2026-10-28-route-and-orchestrate-models-with-github-copilot-fusion/) — GitHub, accessed 2026-10-30
2. [GitHub Copilot Fusion now supports selection of local models](https://github.blog/changelog/2026-10-28-github-copilot-fusion-now-supports-selection-of-local-models/) — GitHub, accessed 2026-10-30
3. [About GitHub Copilot Fusion](https://docs.github.com/copilot/concepts/models/copilot-fusion) — GitHub Docs, accessed 2026-10-30
4. [Enterprise managed settings](https://docs.github.com/copilot/reference/enterprise-administrators/enterprise-managed-settings) — GitHub Docs, accessed 2026-10-30
5. [Managing availability of default models for your enterprise](https://docs.github.com/copilot/how-tos/administer-copilot/manage-for-enterprise/manage-availability-of-default-models) — GitHub Docs, accessed 2026-10-30
6. [Project HydraFusion: frontier quality via multi-model orchestration](https://github.blog/ai-and-ml/github-copilot/project-hydrafusion-frontier-quality-via-multi-model-orchestration/) — GitHub, accessed 2026-10-30
7. [GitHub Copilot in JetBrains Universe releases](https://github.blog/changelog/2026-10-28-github-copilot-in-jetbrains-universe-releases/) — GitHub, accessed 2026-10-30