Route and orchestrate models with GitHub Copilot Fusion #211017
Replies: 7 comments 3 replies
|
I am the administrator of a GitHub copilot business plan, and we do not seem to have access to the Fusion feature that has just been rolled out. I can see no mention in the settings either, to disable or enable or anything like that. Does not appear in the VSCode Copilot chat model picker at all, either. |
|
Copilot Fusion option is still not available. I have a Pro plan with VS Code Version: 1.144.0. |
in another chat...
Not sure exactly when it will pick Opus 5.5, which is what I'd pick for the high complexity task (we have it enabled on our account). |
|
Feature request: allow Fusion Max without GPT-6 Astra On our Copilot Business plan, Fusion Max shows as Unavailable because GPT-6 Astra and GPT-6 Sol aren't enabled. We already allow Claude Opus 5.5 and GPT-6.1 Sol. We'd like to use Fusion Max's orchestration patterns (cascade and critique) with those models. Astra costs 2.5× Opus 5.5 per token (5× for cached input), so enabling it only to unlock Max is a difficult cost decision. Today Max is all-or-nothing: if any required model is disabled, the whole level is unavailable. It would help to have either of these:
Related to @swrobel's comment: it would also help to see when Fusion will escalate to Opus 5.5. |
|
Sometimes selects 6.1 Sol , sometimes 6 Luna but never 6.1 Terra, which seems like it should be useful / optimal in some cases |
|
Some months ago, you introduced the Efficiency, Balanced, and Intelligence tiers for Auto model selection in the Local harness. This let me select Efficiency for simple, well-scoped tasks, which in practice routed them to Haiku/Luna-class models and typically consumed only a small number of credits—usually single digits. With Fusion, that level of control is no longer available to me as a Copilot Pro user. Fusion Standard is the only tier available on my plan, and while I understand that it is intended to balance cost, quality, and latency for everyday work, it gives me no practical way to constrain the models it may select or the resulting credit consumption. In my experience, it has used tens of credits for tasks I expected to cost single digits. For example, I cannot tell whether a straightforward documentation update will be handled by a Luna-class model for a fraction of a credit or by a Sonnet-class model for substantially more. That uncertainty makes me reluctant to use Fusion for routine work. In my view, Fusion is currently designed primarily for the high end of the spectrum: complex agentic tasks where routing and multi-model orchestration can add value. It is not well suited to the low end, where the priority is predictable, inexpensive handling of small edits, documentation, formatting, and routine infrastructure work. As context, I recently used the Local harness’s Efficiency tier very successfully for homelab IaC automation and small Ansible module development. It did introduce some documentation drift—for example, a state document gradually became a changelog—but that was easy to manage with targeted prompts and occasional reviews using the Balanced tier. I do not believe I could achieve the same cost-efficient workflow with Fusion’s current model-selection behavior. So I am back to manually selecting models, which requires me to keep track of model availability, capabilities, and pricing—precisely the work Fusion is intended to remove. Please consider adding a lower-cost Fusion effort level, comparable to the former Efficiency tier, with:
Without something like this, I will need to continue using the Local harness for as long as it remains available, or use manual model selection rather than Fusion. That is disappointing, because the tier-based Auto routing in the Local harness worked very well for this use case, and it is now being deprecated after only a few months. |
|
@ebndev Thanks for the detailed write-up, this is a useful use case. Two things from the docs that might help a bit: you're billed at each used model's standard rate with no separate Fusion charge (paid plans also get 30% off model costs with Fusion through Nov 30), and you can hover over the effort level name under a response to see which models were used, so you can check what a docs update actually routed to. I couldn't find anything in the docs about limiting Fusion to a smaller model pool or setting a per-prompt credit ceiling, so I don't know of an existing setting for what you're asking. |

Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
GitHub Copilot Fusion is generally available. Instead of picking a model for each task, choose Copilot Fusion and let Copilot intelligently route and orchestrate which models it uses for you.
Copilot Fusion brings together Project HydraFusion's multi-model orchestration and Copilot auto model selection's fine-tuned model selection, leveraging the benefits of both to balance performance and cost efficiency.
Since Project HydraFusion launched, we've improved its performance on the DeepSWE benchmark. Using the GitHub Copilot harness, Fusion High and Max delivered task completion scores up to 8.7 points higher than equivalent effort levels from the leading frontier models, while reducing estimated costs by up to 77%.
Smarter choices for developers
When you select Copilot Fusion in the model picker, you aren't selecting a specific model. You're choosing a system that routes and orchestrates models for each task. There's no need to switch models during a task or decide which model suits each step. For each prompt, Copilot Fusion determines which models to use and how they work together, while you choose the right level of effort.
To use it, select Copilot Fusion in the model picker, then choose the effort level.
Watch as Copilot Fusion analyzes each task and selects a model to complete it, orchestrating multiple models through the following workflows when needed.
More control for administrators
Administrators control which models Copilot Fusion can use and whether Copilot Fusion is the default.
Billing
When you use Copilot Fusion you're billed for each model it uses, at each model’s standard rate. For a limited time, users on paid Copilot plans will receive a 30% discount on model costs when using Copilot Fusion.
Get started
Copilot Fusion is available on all Copilot plans in Visual Studio Code, the GitHub Copilot app, GitHub Copilot CLI, and JetBrains IDEs. Support for additional clients is coming soon.
Update your client to try it out. Open the model picker, select Copilot Fusion, and choose an effort level.
🌟 Leave a comment!
To learn more, see our documentation on GitHub Copilot Fusion and join the discussion by leaving feedback in the comments below!
Tune in for Day 2 of GitHub Universe & check out our round-up post of releases in the Community that need your input ⏭️
All reactions