The news lacks technical details or a timeline, but confirmation comes from a heavyweight: according to Alexandr Wang, CEO of Scale AI, interviewed by CNBC, Meta is working on an open source variant of Muse Spark, the coding assistant developed by Anthropic. A clear signal that Mark Zuckerberg's company aims to conquer the AI developer tools market, challenging not only Anthropic but also OpenAI and its Codex.

The name Muse Spark is synonymous with a proprietary approach to assisted coding: a model accessible via API, integrated into development environments, but one that forces companies to share portions of code with third-party servers. It's no surprise that Meta, a long-time champion of open source in the LLM space with its Llama family, wants to offer an alternative. An open-source code LLM would change the game: it would enable self-hosted deployment, fine-tuning on internal codebases, and optimization for enterprise hardware, from edge to on-premise data centers.

For businesses, the difference is significant. In sectors like defense, finance, or healthcare, software development pipelines handle code that must remain under sovereign control. An open source model can be run locally, subjected to quantization to reduce VRAM usage, and tailored to specific needs without ever exposing code tokens outside the corporate perimeter. Inference architectures can scale from workstations with consumer GPUs to multi-GPU servers, and the TCO, while higher than a simple API subscription in terms of upfront investment, pays off with the certainty that no data leaves the organizational boundary.

Of course, the devil is in the still-unknown details. What will the base model be? How large? Will it be a pre-trained checkpoint from Meta itself or a derivative of Llama? And above all, will it be released under a truly open license, usable even in commercial contexts without restrictions? If Meta follows the path traced with previous models, a permissive or at least community-friendly license can be imagined. But until we see a repository and a technical spec sheet, everything remains speculation.

The fact remains that Meta's move is an important temperature check. Open source is permeating every layer of software development, and coding assistants are no exception. Just as Llama paved the way for self-hosted enterprise chatbots, a potential «Code Llama» would further broaden the range of use cases for those who want to do without cloud APIs. In the meantime, for those evaluating the adoption of an AI code assistant today, the possibility of an imminent open source alternative might suggest postponing lock-in decisions on proprietary services. It's a project worth keeping an eye on.