GitHub Copilot Fusion Explained: How It Picks Your AI Model
Prefer to watch? Watch on YouTube · Subscribe for new builds · Read the transcript
GitHub Copilot can now pick the AI model for you. The feature is called Fusion, and this short video explains how it decides which model, or models, work on your task. It's for anyone who uses AI coding tools and doesn't want to keep guessing which model to choose.
What I cover in this video
- 0:00 What Fusion is: Copilot choosing the model for you
- 0:00 Why you pick an effort level now, not a model
- 0:00 Workflow one: Single, where one model solves the task directly
- 0:00 Workflow two: Cascade, where a draft gets accepted or escalated
- 0:00 Workflow three: Critique, which goes draft, review, revise
- 0:00 How Fusion Max compares to Claude Code with Opus 5.5 on an independent coding index
You choose an effort level, not a model
Normally you open a model picker in a coding tool and have to decide which model fits the job. Fusion changes that. You pick an effort level instead:
- Standard
- High
- Max
Fusion then reads your task and picks the lightest workflow that still hits the quality bar. Easy tasks don't get the heaviest setup, and harder tasks get more models working on them.
The three workflows Fusion picks from
1. Single
One model solves the task directly. This is the simplest path and the one you'd expect for simple tasks.
2. Cascade
An efficient model writes a draft first. Then a quality check looks at it and does one of two things:
- Accepts the draft if it's good enough, or
- Escalates to a stronger model if it isn't.
So you only bring in the stronger model when the cheaper one falls short.
3. Critique
Here two models work together:
- One model drafts the solution.
- Another model reviews it.
- The first model revises based on that review.
This is the most involved of the three, because you get a second opinion built into the process.
How Fusion Max scores, and what it costs
On Artificial Analysis' Coding Agent Index, which is an independent benchmark, Fusion Max scored 60. That's within 2 points of Claude Code with Opus 5.5 Max, at about half the cost per task.
There's one catch to know about: you pay for each model Fusion uses. A Cascade that escalates, or a Critique with two models, means more than one model is part of the bill for that task. Even so, by that index the overall cost per task came out at roughly half.
That leaves the question I end the video with: would you let it pick for you? If you're comparing coding tools, my best vibe coding tools page and the AI agents hub are good places to keep going.
Key takeaways
- GitHub Copilot Fusion picks the AI model for you based on your task.
- You choose an effort level (Standard, High or Max) instead of a specific model.
- Fusion picks the lightest workflow that meets the quality bar: Single, Cascade or Critique.
- Fusion Max scored 60 on Artificial Analysis' Coding Agent Index, within 2 points of Claude Code with Opus 5.5 Max, at about half the cost per task.
- You pay for every model Fusion uses on a task.
FAQ
What is GitHub Copilot Fusion?
Fusion is a GitHub Copilot feature that picks the AI model for you. You set an effort level, and it reads your task and picks the lightest workflow that still meets the quality bar.
What are the Copilot Fusion effort levels?
There are three: Standard, High and Max. You choose one of these instead of choosing a specific model.
What is the difference between Cascade and Critique in Copilot Fusion?
In Cascade, an efficient model writes a draft and a quality check either accepts it or escalates to a stronger model. In Critique, one model drafts, a second model reviews it, and the first model revises.
How does Copilot Fusion Max compare to Claude Code?
On Artificial Analysis' Coding Agent Index, Fusion Max scored 60, within 2 points of Claude Code with Opus 5.5 Max, at about half the cost per task. You pay for each model Fusion uses.
Transcript
Read the full transcript
0:00GitHub Copilot can now pick the AI model for you. It's called Fusion. Here's how it works. You don't choose a model, you choose an effort level. Fusion reads your task, then picks the lightest workflow that still hits the quality bar. Workflow one, a single model solves it directly. Workflow two, cascade. An efficient model writes a draft. A quality check accepts it or escalates to a stronger model. Workflow three, critique. One model drafts, another reviews it, and the first model revises. On an independent coding index, Fusion Max scored within two points of Claude Code with Opus 5.5 at about half the cost. Would you let it pick for you?
Ready to start building with AI?
Join the Shipping Skool community free. Paid members get live bootcamp calls every week and weekly-updated courses.
Join Shipping SkoolKeep reading
Claude Opus 4.7 Benchmarks: What It Means for OpenClaw Agents
Claude Opus 4.7 beats Opus 4.6 and GPT-5.4 on SWE-bench Verified (87.6%), SWE-bench Pro (64.3%), and MCP-Atlas tool use (77%). It also checks its own work before it answers. The price didn't drop, so I save it for the two or three highest-stakes jobs in an OpenClaw pipeline and leave the rest on cheaper models.
Awesome Codex Subagents: 136 Specialist AI Coding Agents Explained
VoltAgent's awesome-codex-subagents repo has 136 specialist Codex subagents in 10 categories. Each one is a TOML config with its own instructions, model routing, and sandbox permissions. You clone the repo, copy the TOML files into your Codex agents folder, and tell Codex which specialist to use for each task.
Claude Fable 5.2 Explained: Pricing, Benchmarks, and Effort Levels
Anthropic's Claude Fable 5.2 costs $8 in and $40 out per million tokens, down from $10 and $50. At High effort it roughly matches Opus 5.5 at Max on FrontierCode for about half the cost per task. I'm keeping Opus 5.5 as my daily driver and using Fable for big, hard projects.
