GitHub Copilot in VS Code: Configurable Thinking Effort for Reasoning Models
GitHub CopilotGitHub Copilot in VS Code now lets developers control how deeply reasoning models think before responding, directly from the model picker. In VS Code v1.113, models that support configurable reasoning — including Claude Sonnet 4.6 and GPT-5.4 — expose a Thinking Effort submenu (Low, Medium, High) without requiring navigation to settings. VS Code retains the selected effort level per model across conversations, giving developers a quick way to balance response depth against speed for each task.
Sources & Mentions
2 external resources covering this update
Configurable Thinking Effort: Tune Reasoning Depth Per Request
Starting with VS Code v1.113, GitHub Copilot exposes a Thinking Effort control directly in the model picker for reasoning-capable models. This gives developers a simple, accessible way to adjust how much computational reasoning a model applies to each request — without navigating through VS Code settings.
How It Works
When a reasoning model like Claude Sonnet 4.6 or GPT-5.4 is selected, a "Thinking Effort" submenu appears directly in the model picker dropdown. Developers choose from three levels:
- Low — Faster responses with lighter reasoning, suitable for straightforward tasks.
- Medium — Balanced reasoning depth for most development tasks.
- High — Maximum reasoning for complex, multi-step problems where depth matters more than speed.
VS Code persists the selected effort level per model across conversations, so developers do not have to reconfigure it for each new chat session.
Why It Matters
Not every coding task benefits from heavy reasoning. Debugging a simple typo does not warrant the same computational depth as designing a complex algorithm or architecting a new system. Configurable thinking effort lets developers consciously trade off speed and depth based on what the task actually demands — and do so in a single click from the UI.