Faster responses, fewer rate limits, and 10% off premium requests. We are launching auto, now available in Visual Studio Code! Auto picks the right model for the job in GitHub Copilot, chosen automatically based on capacity and performance. Great products these days are rarely powered by a single model. They rely on a system of models working together, each playing to its strengths. With auto, we are bringing that ensemble approach directly to customers. Over time, auto will get smarter. It will dynamically scale between small and large models based on task complexity, and fold model selection into the development loop itself. 🔗 https://lnkd.in/gTHA-ZsR
Amazing to see this new feature Asha Sharma
Each time I try Auto in Cursor, it more or less, disappoints... When time is money you want to choose the model you can trust most!
This is exciting
Great to see!!
Sounds like a smart, efficient way to optimize resource usage and improve the developer experience! Excited to see how Auto evolves.
Hey Asha. I have a question about your post. You used the word "ensemble". Is the auto feature in Visual Studio Code truly an ensemble where it does something like bagging or boosting, ultimately returning what is evaluated as the best result from multiple results across multiple models? Or is it "guessing" what it thinks is the best model to use based on the input prompt, and then ultimately routing?
Okay this is a cool functionality—so many of our customers have been asking for this. Thank you team!
Program Management | Client Solutions Delivery | MS in Engineering Management
1wGame-changing move with GitHub Copilot’s ‘auto’—smart ensemble of LLMs (e.g., Claude Sonnet 4, GPT-5, mini) auto-selects for speed, capacity, cutting rate limits and offering 10% premium discounts. It evolves from single-model limits, mirroring cloud autoscaling for AI, potentially easing dev fatigue by 20-30% via adaptive orchestration. For bolder strides: Enable real-time intra-session model switching via on-device profiling (code context analysis), with “why this model?” tooltips and user overrides—could slash costs 40% in mixed workloads, enhancing transparency and domain fits like UI vision tasks. Pumped for the future! #AICoding #GitHubCopilot #DevTools