Copilot Fusion GA: Multi-Model Routing Behind One Picker

Introduction

On October 28, 2026, GitHub announced in its changelog that GitHub Copilot Fusion is generally available. Instead of picking a model for each task, a developer selects Copilot Fusion in the model picker, chooses an effort level, and lets Copilot decide which models to use and how they work together for each prompt. GitHub says Fusion brings together the multi-model orchestration of Project HydraFusion, its earlier research preview, and the fine-tuned model selection of Copilot auto model selection, with the aim of balancing performance and cost efficiency. The release was announced during GitHub Universe 2026, where GitHub also showed other agentic-delivery products; this review covers Fusion only.

Feature Overview

Effort levels. Selecting Fusion is not selecting a model. The developer picks one of three levels, and GitHub describes what each one prioritizes.

The changelog uses the word "prioritizes" and does not list which specific models sit behind each level.

Orchestration workflows. For each prompt, Fusion determines which models to use and how they work together. The changelog names three workflows it uses when needed.

Reported benchmark. GitHub reports that in the Artificial Analysis Coding Agent Index, Copilot Fusion Max scored 60, within two points of Claude Code with Opus 5.5 Max. According to GitHub, it reached that result at about half the average API cost per task, using about half as many tokens and taking about 40% less time per task. These are GitHub's own figures, published in its announcement, and should be read as vendor-reported until independently reproduced.

Administrator controls. Fusion only uses models that are enabled in an enterprise's and organization's model policies. Enterprise administrators can view which models Fusion picks from, set each model's effective policy state independently, and disable a model for Fusion without disabling it for everyone. They can also set Fusion, at any effort level, as the default model for Copilot users through Enterprise managed settings.

Billing and availability. Users are billed for each model Fusion uses, at that model's standard rate. For a limited time, users on paid Copilot plans receive a 30% discount on model costs when using Fusion. Fusion is available on all Copilot plans in Visual Studio Code, the GitHub Copilot app, GitHub Copilot CLI and JetBrains IDEs, with support for additional clients described as coming soon. Research-preview participants who enabled auto, HydraFusion or HydraFusion Max are moved automatically to the corresponding Fusion effort levels.

Usability Analysis

For an individual developer, the change is mostly about removing a decision. The changelog says there is no need to switch models during a task or to decide which model suits each step; the developer only chooses how much effort the work deserves. That maps naturally to a daily pattern: Standard for routine edits, High when a task needs deeper reasoning, Max for the hardest work.

AT&T's Hari Pokhrel, a Principal Software Engineer, is quoted in the changelog saying Fusion Standard improved token efficiency across AT&T by selecting the right model automatically. He describes a habit Fusion is meant to remove: developers choosing a more powerful model for a complex task and forgetting to switch back for everyday work. This is a customer testimonial published by GitHub, not an independent measurement.

Because billing follows every model Fusion uses, the cost of a task depends on the route taken. A Cascade run that accepts the first draft never calls the stronger model, while a Critique run always involves a drafting model and a reviewing model. The changelog does not give a per-workflow cost breakdown, so the benchmark's cost claim is the main published evidence on how this plays out.

For administrators, the controls are the notable part. Limiting Fusion to approved models and disabling a model for Fusion alone lets an organization adopt routing without widening what it allows elsewhere.

Pros and Cons

On the positive side, a single picker entry replaces per-task model switching, admins keep model-level policy control, and GitHub reports near-parity benchmark results at lower cost. On the limiting side, the changelog does not say which models each level uses, the 30% discount is time-limited with no stated end date, and the headline benchmark is vendor-reported against one named configuration.

Outlook

Several items are stated as pending. Support for additional clients is listed as coming soon, and the discount is described only as limited-time. Details the changelog leaves out include the model lineup per effort level, how the quality check in Cascade decides to escalate, and how long the promotional pricing lasts. Independent benchmarking of Fusion against single-model setups would show whether the reported cost and time advantages hold across different codebases. At the product level, GitHub's Universe post frames the idea as developers not having to manage every model choice to get the right result at the right price.

Conclusion

Copilot Fusion turns model choice into an effort setting and moves routing into the product, with enterprise policy controls around it. It suits teams that use several models and want less manual switching, and administrators who need to keep model access governed. The benchmark and customer claims come from GitHub and AT&T through GitHub's own channel, and the opaque per-level model lineup is the main thing to weigh before relying on it for cost planning. Rating: 4 out of 5.

Editor's Verdict

Copilot Fusion GA: Multi-Model Routing Behind One Picker earns a solid recommendation within the AI tools space.

The strongest case for paying attention: a single picker entry replaces per-task model switching, so developers choose an effort level rather than a model. That alone raises the bar for what readers should expect in this space. Reinforcing that, administrator controls limit Fusion to approved models, allow per-model disabling for Fusion only, and let Fusion be set as the default — practical value rather than just headline appeal. The broader signal worth registering is straightforward: the unit of choice shifts from a specific model to an effort level, because selecting Fusion in the picker hands model selection for each prompt to Copilot. On the other side of the ledger, one constraint is real rather than a marketing footnote: the changelog does not say which specific models each effort level uses, so routing is harder to predict from the announcement alone. It should factor into any serious decision. Layered on top of that, a 30% discount applies only for a limited time, and no end date is given, which complicates longer-term cost planning — which narrows the set of teams for whom this is an obvious yes.

For engineering teams already juggling several models in Copilot, and for administrators who need model access to stay governed, this is a serious evaluation candidate, not just a curiosity to bookmark.