TL;DR: GitHub made GitHub Copilot Fusion generally available on October 28, 2026, at its Universe conference in San Francisco. Instead of picking a model yourself, you pick Copilot Fusion plus an effort level (Standard, High or Max), and Copilot decides which models to use and whether one model should draft while another checks the work. GitHub says Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within two points of Claude Code with Opus 5.5 Max, at about half the average API cost per task. Those figures are GitHub’s own. You pay each underlying model’s standard rate, with a limited-time 30% discount on paid plans.
GitHub Copilot Fusion is GitHub’s answer to a problem most developers who use AI coding tools know well: choosing a model for each task, then forgetting to switch back. Fusion turns model choice into a background decision. Below is what GitHub announced, how the three execution patterns work, how billing works and what administrators can control.
What GitHub announced
According to the official GitHub changelog, Copilot Fusion combines two earlier features:
- Project HydraFusion, a multi-model orchestration system GitHub released as a research preview on September 4, 2026.
- Copilot auto model selection, which already matched a prompt to a suitable model.
When you choose Copilot Fusion in the model picker, GitHub’s documentation says you “aren’t selecting a specific model.” You are choosing a router that looks at each prompt and decides which models run it. The New Stack reports that Fusion launched alongside other Universe announcements, including the GitHub Copilot app, Copilot Projects and Copilot Automations.
The three effort levels
Developers keep one decision for themselves: how hard Copilot should try. GitHub describes three levels:
| Effort level | What GitHub says it prioritizes | Typical use |
|---|---|---|
| Fusion Standard | Balance of cost, quality and latency; one model per task | Everyday coding |
| Fusion High | More capable models, moderate token savings | Complex bug fixes, changes across several files |
| Fusion Max | Frontier capability while still watching token use | The hardest problems that need deep reasoning |
If you used the research preview, GitHub says earlier selections of auto, HydraFusion or HydraFusion Max move automatically to the matching Fusion effort level.
How Copilot Fusion routes a task
For each prompt, Fusion weighs what GitHub calls capability signals for reasoning, code generation, debugging and tool use, then picks one of three execution patterns.
Single
One model answers the prompt directly. GitHub says this behaves like a normal request to one model, so simple tasks don’t pay for extra review steps.
Cascade
An efficient model drafts a solution. A judge then checks the draft and either accepts it or escalates to a stronger model. The idea is that cheap models handle what they can, and expensive ones only step in when needed.
Critique
One model drafts, several other models review it in parallel, their feedback is combined, and the original model revises. The HydraFusion preview post said the critic comes from a different model family, which is meant to catch mistakes a single model might repeat.
The pattern is chosen per prompt, so one session can mix them: a quick rename might run as Single, while a follow-up refactor runs as Critique. GitHub also says Fusion considers real-time model availability and tries to switch models only at “natural cache boundaries” to avoid extra caching costs. You can watch which model handles each step, but only the final answer is shown, since intermediate drafts may be thrown away.
One practical detail: Fusion has no context window of its own. Each step runs within the limits of the model doing it, and the context window shown for Fusion is a conservative value based on the smallest model in its pool. If you’re unsure what that means, our explainer on context windows in AI covers it.
The benchmark claims, in context
GitHub’s headline claim is that Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, “within two points of Claude Code with Opus 5.5 Max,” while using about half the average API cost per task, about half as many tokens and about 40% less time per task. These are GitHub’s numbers and haven’t been independently reproduced yet. “Within two points” also means Fusion Max still scored slightly below the top result it is compared with; the pitch is efficiency, not a new record.
GitHub also published a quote from Hari Pokhrel, a principal software engineer at AT&T, who said Fusion Standard “improved token efficiency across AT&T” because “developers would often choose a more powerful model for a complex task and forget to switch back for everyday activities.” That is one customer’s experience supplied by GitHub, not a broad study.
Pricing and availability
- Billing: you’re charged for each model Fusion uses, at that model’s standard rate. Cascade and Critique cost more than Single because they add model calls.
- Launch discount: for a limited time, paid Copilot plans get 30% off model costs when using Fusion.
- Plans: available on all Copilot plans.
- Where it works: Visual Studio Code, the GitHub Copilot app, GitHub Copilot CLI and JetBrains IDEs, with more clients “coming soon.”
To try it, update your client, open the model picker, choose Copilot Fusion and pick an effort level.
What administrators can control
Fusion only uses models enabled in an enterprise’s and organization’s model policies. Enterprise admins can see which models Fusion picks from, disable a model for Fusion without disabling it for everyone, and set Fusion at any effort level as the default for Copilot users through Enterprise managed settings. For teams with rules about which AI providers can see their code, that matters more than the benchmark score.
Why it matters
Model routing is becoming standard across coding tools. The New Stack notes that Ramp moved model selection into its open-source Open SWE agent this month and Replit made its own routing system the default earlier this year. GitHub’s version stands out because of its reach and because it combines routing with multi-model review in one picker option.
The trade-off is control. Developers give up knowing in advance which model will touch their code, and costs vary by task instead of by fixed model. For cost-conscious teams, the 30% launch discount and the AT&T example make it worth testing on real work before deciding. It also fits a wider trend toward agentic AI, where systems plan and check their own steps instead of answering in one shot.
Bottom line
GitHub Copilot Fusion swaps the model picker for an effort dial. If GitHub’s efficiency numbers hold up outside its own tests, it could cut AI coding bills without much loss in quality. Until independent results arrive, treat the benchmark as GitHub’s claim, start with Fusion Standard on everyday tasks, and compare your usage costs against your current fixed-model setup. For background on the models Fusion routes between, see our guides to large language models and reasoning models.
The Teck Hustlers Editorial Team covers technology, business, finance and gaming news for readers who want the story behind the headline. Every article is researched, written and edited by our in-house team, checked against company announcements, filings and reputable reporting, and updated when new facts emerge. We focus on what each development means for founders, professionals, investors and everyday users. Have a tip or a correction? Reach us through our contact page.
