GitHub Copilot goes multimodel, adding support for Google’s Gemini and Anthropic’s Claude LLMs – SiliconANGLE

Microsoft Corp.’s GitHub sprung a surprise today as it revealed that its popular generative artificial intelligence coding assistant, GitHub Copilot, will no longer be powered exclusively by OpenAI’s GPT models.

Instead, the company said at its annual GitHub Universe event that it’s adopting a multimodel approach, giving developers the opportunity to tap models from rival companies such as Anthropic PBC and Google LLC too.

The revelation came alongside a host of other announcements, with GitHub launching a new AI tool called Spark that can automate much of the work that goes into building web applications, and more general updates to GitHub Copilot in VS Code and Xcode.

However, it’s the decision to make GitHub Copilot a multimodel tool that is by far the biggest announcement. In a blog post, GitHub Chief Executive Thomas Dohmke said users can now use Anthropic’s Claude 3.5 Sonnet large language model in Copilot Chat’s web and VS Code interfaces, with Google’s Gemini 1.5 Pro to arrive as another option in the coming weeks.

In addition, GitHub Copilot will support an expanded range of OpenAI models, including more specialist ones such as GPT o1-preview and GPT o1-mini, which are said to be more focused on advanced reasoning than GPT-4o, which is the model it currently uses exclusively.

The idea is that developers will be able to switch between various models based on whatever it is they’re trying to achieve, and is an acknowledgment that, in the AI world, there are different horses for different courses, with some models better suited than others, depending on the task.

The choice won’t just be up to developers.. Organizations will also be able to select which models they’re willing to make available to various team members using GitHub Copilot Enterprise.

Dohmke said the approach makes sense because it has become clear that there is no one model to rule every scenario. “The next phase of AI code generation will not only be defined by multimodel functionality, but by multimodel choice,” he wrote.

Anthropic and Gemini will first become available in the GitHub Copilot web and VS Code chat interfaces, but the company eventually plans to bring them to every one of its surface areas and functions, Dohmke said, including the Copilot Workspace, GitHub’s command line interface and various other tools.

A multimodel future?

Given Microsoft’s influence over GitHub, the move has led to speculation that the Redmond-based company itself may also decide to end its exclusivity arrangement with OpenAI and enable users to access other models in its own Copilot tools.

While GitHub kicked off the copilot craze when it debuted its generative AI coding assistant in 2021, Microsoft has introduced a host of its own Copilots across platforms such as Windows and Office.

As the biggest backer of OpenAI, Microsoft has taken full advantage of that relationship, using its GPT models as the engine behind its various Copilot tools, and has never given any indication that it’s unhappy with their performance. But with GitHub’s switch to a multimodel approach, it’s likely that Microsoft has at least considered doing the same.

Whether or not it will actually do so is far from certain. Microsoft acquired GitHub back in 2018, but unlike most other acquisitions it has allowed the company to operate more or less independently since then. So the same thinking prevalent at GitHub may not be present at Microsoft.

For a long time, Microsoft sang the praises of OpenAI’s GPT models, pitching them as the best of the best so that customers would think its exclusive access gives it an advantage over its competitors.

However, it has become clear to most that OpenAI’s models really aren’t that superior at all, with Google’s Gemini and Anthropic’s Claude models both consistently demonstrating some impressive capabilities of their own. With that, the competition in the AI industry has shifted, and Microsoft is more concerned with providing enterprises with the tools and platforms they need to build generative AI applications. And one thing its customers want is flexibility around model choice, which is why the Azure AI service now offers multiple different LLMs from a variety of vendors.

Then again, the choice of which model to use is less pronounced in some applications than others. The intricacies of writing code mean that GitHub Copilot can definitely benefit from having greater choice, as some models are more proficient at specific programming languages than others. But that may not be the case for Copilots tasked with writing newsletters or fixing user’s grammar. So it’s not entirely sure if Microsoft will curtail OpenAI’s exclusivity advantage just yet.

Holger Mueller, an analyst with Constellation Research Inc., told SiliconANGLE he thinks Microsoft really has no choice but to abandon its exclusive arrangement with OpenAI and embrace a multi-model future.

“There’s no single model that can be described as ‘the best’ for every programming language, and it’s an area that’s seeing fierce competition from dozens of algorithm vendors,” the analyst said. “Which model is considered best is often a very subjective thing, and each developer will have their own personal preference. So Microsoft needs to provide options, as that’s what will keep developers happy and increase their velocity.”

GitHub Spark: Building apps with words

While the tech world watches with baited breath to see if Microsoft does follow GitHub’s lead, there were plenty of other announcements at the GitHub Universe event. Perhaps the most significant is the launch of Spark, which is an AI tool for building web apps using natural language.

In a demonstration, GitHub showed how an initial prompt – essentially just a description of the app – generates a series of live previews of what that app could potentially look like. Users can compare the different previews, select the one they like, and then enter further prompts to change the app’s look and feel.

Of course, experienced developers will still be able to manipulate the code if they desire, while inexperienced users will just be able to continue prompting it with their natural language until they get the design they want.

Once the user is happy with the app, GitHub Spark can then deploy it wherever the user wants it, on a desktop, tablet or smartphone, for example. They’ll also be able to share their app with others, so their colleagues can tweak it if they desire.

According to GitHub, Spark will help the company to fulfill its vision of creating 1 billion developers in the world.

“For too long, there has been an unscalable barrier of entry separating a vast majority of the world’s population from building software,” Dohmke said. “With Spark, we will enable over 1 billion personal computer and mobile phone users to build and share their own micro apps directly on GitHub.”

Featured image: GitHub