GitHub has announced today that developers can now switch between various large language models when using Copilot Chat, its code-focused service similar to ChatGPT. Rather than being solely powered by OpenAI’s GPT-4, developers can now select from Anthropic’s Claude 3.5 Sonnet, Google’s Gemini 1.5 Pro, and OpenAI’s GPT-4o, o1-preview, and o1-mini for Copilot Chat. Anthropic will be the first to launch, followed by Gemini in a few weeks.
Some might interpret this as Microsoft distancing itself from OpenAI, but GitHub CEO Thomas Dohmke sees it as offering developers more options. Not all models are equally efficient for every development task, and some are better suited for certain languages than others.
“We believe that relying on a single model is a thing of the past,” Dohmke explained. “GitHub Copilot already utilizes different versions of GPT-3 or -4 models for various tasks. We used smaller, faster models for low latency needs and bigger, more complex models for high accuracy demands. Here at GitHub, we value developer choice. Developers may have different preferences based on company policies, benchmarks they’ve encountered, programming languages, personal preferences, or previous model usage for other scenarios. Hence, we are now officially collaborating with Anthropic and Google.”
The groundwork for this transition was initiated when developers were first given the option between GPT-4 and GPT-4 o1, which was introduced just over a month ago. Currently, developers can choose models only for Copilot Chat and the new Spark feature, but GitHub plans to extend this option to all its tools in the future.
According to Google Cloud CEO Thomas Kurian, developers want a diverse range of models that are tailored for code tasks like generation, refactoring, and optimization. Gemini models are particularly skilled in these areas and are now available on popular developer platforms and environments, including GitHub Copilot, enabling millions of developers worldwide to leverage enterprise-grade AI from Google Cloud.
