Meta is preparing to launch Code LLaMA, a model designed to generate programming code. The company is positioning it as an open-source alternative to tools from OpenAI, Google and others, with potential appeal for businesses that want to build coding assistants around their own needs.
A model aimed at coding assistants
Code LLaMA builds on Meta’s recently released LLaMA 2, a large language model that can understand and generate conversational text. The new model is intended to help developers create AI assistants that suggest code automatically.
That could give organizations another option alongside paid coding assistants. Microsoft’s GitHub Copilot is based on OpenAI’s Codex, while Code LLaMA is presented as a model companies could use to develop their own tools.
For businesses, the attraction may be the ability to tailor a coding assistant to their software and workflows. The source article notes that larger companies may favor developing their own assistant to protect their source code.
Why companies may pay attention
AI coding tools already have a business market. Microsoft CEO Satya Nadella said 27,000 companies were paying for an enterprise license for GitHub Copilot. The article also points to other alternatives and the use of GPT-4 with a code interpreter for coding-related work.
Code LLaMA could widen that field by making a code-generating model available for organizations to adapt. Tim Chen, a managing partner at Essence VC, told The Information that enterprise adoption could accelerate if companies can modify a model for their own use. He gave Bank of America as an example of a company that might adapt a model to generate code based on its existing code base.
The practical promise is flexibility: a company could build a coding assistant with behavior suited to its code and internal requirements. That possibility is distinct from simply subscribing to a ready-made assistant, though the source does not detail how Code LLaMA would be deployed or customized.
“Open source” still needs a closer look
The label “open-source” does not by itself settle what users will be allowed to do with Code LLaMA. The definition of open source in AI is being debated, and the article raises the question of how open Meta’s model will be.
LLaMA 2 has been marketed as open-source, but its use comes with restrictions and a limit on the number of users allowed for a project. The source says commercial use is permitted up to 700 million active users per month. That allowance is large, but the restrictions mean organizations still need to understand the terms attached to the model.
For a company weighing a coding assistant, licensing is part of the decision alongside technical fit. The ability to adapt a model may matter, but so does whether its conditions work for the organization’s intended use.
What Code LLaMA could change
If Code LLaMA delivers on its stated role, it could give developers another foundation for AI code suggestions and give companies more room to shape their own assistant. Its arrival would add to a market that already includes GitHub Copilot, other alternatives and GPT-4 used with a code interpreter.
The key questions are how the model performs for real coding tasks and what its final access terms allow. The source describes a planned launch and a potential rival to proprietary coding software; it does not provide results or details that resolve those questions.