GitHub Copilot in Visual Studio Adds Thinking Effort Controls
GitHub CopilotGitHub Copilot in Visual Studio now lets developers choose Low, Medium, or High thinking effort for supported models, trading reasoning depth for token consumption on a per-task basis. The update also redesigns the model picker so developers can pin frequently used models, collapse rarely used ones, and open an expanded Manage Models view showing cost, capabilities, and context window size for every connected Copilot and custom model.
Key Takeaways
- Three effort levels — Low, Medium, High — let developers directly control how deeply a Copilot model reasons before responding.
- Low effort minimizes token use, ideal for quick suggestions and simple questions, while High effort maximizes reasoning depth for complex problems.
- The feature is a direct response to usage-based billing, giving developers a lever to manage per-task spend.
- The model picker was redesigned to let developers pin favorite models and collapse rarely used ones.
- A new Manage Models view surfaces cost, capabilities, and context window size for every connected model in one place.
- Thinking effort only applies to models built to support adjustable reasoning, so it won't appear for every model in the picker.
Sources & Mentions
3 external resources covering this update
A Dial for How Hard Copilot Thinks
Visual Studio 2026 version 18.9 introduces thinking effort controls for GitHub Copilot's supported models. Instead of a model always reasoning at a fixed depth, developers can now choose Low, Medium, or High effort directly from the model picker or the expanded Model management view. Low effort favors quick responses with minimal reasoning — well suited to autocomplete-style suggestions and straightforward questions — and uses the fewest tokens. Medium balances depth and speed for everyday coding tasks like explaining a function or writing a unit test. High effort pushes the model to reason more extensively, aimed at tricky algorithms, architecture decisions, and hard-to-debug issues, at the cost of consuming more tokens.
Why It Shipped Now
The timing isn't incidental. GitHub moved Copilot onto usage-based billing earlier in 2026, which ties the cost of a chat or agent interaction directly to token consumption. Thinking effort gives developers a lever to manage that cost themselves: dial reasoning down for routine work to conserve budget, and dial it up only when a problem actually warrants deeper analysis. It's a direct response to the reality that not every prompt needs the same amount of model reasoning behind it.
A Redesigned Model Picker
Alongside thinking effort, Visual Studio 18.9 reworks the Copilot model picker to cut down on clutter. Developers can now pin the models they use most often to the top of the list and collapse the ones they rarely touch. Selecting "Manage Models" opens an expanded view listing every Copilot and custom model connected to the IDE, along with its cost information, capabilities, and context window size — plus controls for managing each one in one place. Combined with a new context-window shortcut that surfaces full Copilot plan usage details and clearer notifications as users approach their usage limits, the update gives developers a much more direct view into what they're spending and where.