GitHub Copilot: Configurable Reasoning Levels for Cloud Agent
GitHub CopilotGitHub Copilot cloud agent now lets developers customize the reasoning level for models that support adjustable reasoning effort, choosing how much time and computation the model spends thinking before it responds. The setting appears as a second dropdown alongside model selection when starting a task, letting developers dial up deeper reasoning for complex architectural or debugging work, or dial it down for simpler tasks to save time and credits. The feature is live in VS Code, Copilot CLI, and the GitHub Copilot app, with more surfaces expected soon, and is available on every paid Copilot plan that supports the cloud agent — Pro, Pro+, Business, Enterprise, and Max.
Key Takeaways
- A new reasoning-level dropdown appears alongside model selection for any model that supports configurable reasoning effort.
- Higher reasoning levels trade speed for quality, giving complex architectural or debugging tasks more deliberation before Copilot responds.
- Lower reasoning levels save time and AI credits, making them a better fit for simple, well-defined tasks.
- The setting is selected once per task and applies for the entire duration of that Copilot cloud agent session.
- Support currently spans VS Code, Copilot CLI, and the GitHub Copilot app, with more surfaces planned.
- The feature is available on every paid Copilot plan with cloud agent access, including Pro, Pro+, Business, Enterprise, and Max.
Reasoning, Tuned to the Task
GitHub Copilot cloud agent now gives developers direct control over how hard a model thinks before it acts. When starting a task with a model that supports configurable reasoning, a second dropdown appears next to the model picker, letting developers select a reasoning level in addition to the model itself. The setting persists for the duration of that task, shaping how much internal deliberation the model performs before producing its plan, code changes, or response.
Why It Matters
Higher reasoning levels can meaningfully improve outcomes on hard problems — untangling a gnarly regression, planning a multi-file refactor, or reasoning through an ambiguous specification — but that extra deliberation comes at a cost. It takes longer to get a result, and it consumes more AI credits and tokens along the way. Previously, developers had no way to trade off between speed and depth beyond picking a different model entirely. Now the reasoning level itself becomes a lever: a quick bug fix can run at a lower reasoning setting to stay fast and cheap, while a gnarly architectural decision can be routed to a higher setting without switching models.
How It Works
The control shows up wherever Copilot cloud agent tasks are started with a reasoning-capable model selected. Developers pick a model first; if that model exposes configurable reasoning, the reasoning-level dropdown appears immediately after it. Copilot cloud agent then carries that setting through the entirety of the task's execution, rather than requiring it to be re-specified at each step.
Availability
The feature is rolling out across GitHub Copilot's cloud agent surfaces, starting with Visual Studio Code, Copilot CLI, and the GitHub Copilot app, with additional entry points expected to gain support soon. It's available to every paid plan that includes cloud agent access: Copilot Pro, Pro+, Business, Enterprise, and Max.