OpenAI o1 reasoning models arrive in GitHub Copilot and Models preview
Developers can now sign up for a preview of OpenAI’s o1-preview and o1-mini models, hosted on Azure, within GitHub Copilot Chat and GitHub Models. The preview targets hands-on evaluation of the new reasoning-focused models in both chat-based coding assistance and standalone model testing.
OpenAI’s o1 series is designed for complex, multi-step reasoning. Unlike previous models focused on quick generation, o1 models work through an internal thought process before responding. In GitHub’s own testing, o1-preview demonstrated a stronger grasp of code constraints and edge cases, leading to more efficient and higher-quality output. The model’s deliberate responses also made it easier to isolate faults and move quickly to fixes.
Swapping models mid-conversation
During the preview, Copilot Chat users in Visual Studio Code can switch the underlying model from the current default (GPT-4o) to either o1-preview or o1-mini. The model toggle is available within a single conversation, allowing developers to move between quick code explanation or boilerplate generation and heavier tasks like designing complex algorithms or investigating logic bugs. This provides a direct comparison of how the reasoning models handle challenging coding scenarios.
Testing in the playground and beyond
Beyond Copilot, both o1 models are accessible in the GitHub Models playground. This environment allows developers to evaluate the models’ distinct capabilities and performance before deciding to integrate them into their own applications.
The preview introduces OpenAI’s latest models to developers working both alongside Copilot and on next-generation LLM-based products.
