Current AI ships the Gap Map for open source AI

Current AI has just published the first version of its Gap Map, an index that documents in detail 421 open source artificial intelligence projects: 266 software tools and libraries, 85 models, 50 datasets and 20 hardware projects, signed by 228 organizations. Alongside those 421 catalogued sits a long tail of 24,400 artifacts still unclassified, which receive no score until someone researches and cites them properly.

The launch is covered by Simon Willison on his weblog, where he admits he is more interested in what the map exposes than in the catalogue itself. And that is the key to the initiative: it is not a directory for showing off how much open source exists, but an attempt to point out where pieces are missing.

What Current AI is

Current AI presents itself as "a global partnership building a public option for AI". It was born as a non-profit at the AI Action Summit in Paris, in February 2025, and it does not start empty-handed: there are 400 million dollars already committed. With that backing, the Gap Map does not look like a weekend experiment but the first public release of an organization with financial muscle and infrastructure ambitions.

Version 0.1, presented this week, splits the catalogued projects into 14 categories spread across three layers of the stack: model components, product and UX, and infrastructure. That structure is what sets it apart from a flat list of repositories. Instead of piling up links, it tries to place each project where it belongs within the ecosystem and, along the way, expose the holes.

Why the name matters

The detail worth noticing is the name itself. It is not called "Open Source AI Map" but Gap Map. The premise is that the value lies not in listing what already exists, but in identifying the areas where open software is weak or simply absent compared to closed alternatives. A dataset nobody maintains, an infrastructure category without any mature free option, a type of model that only has proprietary implementations: those are the voids the project wants to make measurable.

That 24,400 artifacts fall outside the initial classification is not a flaw, it is methodological honesty. The team prefers not to score what it has not reviewed rather than inflate the catalogue with unverified repositories. At a time when "the 100 best AI tools" lists generated without criteria abound, an index that admits what it has not yet looked at is unusual.

Who it is useful for

For anyone building on open source, the Gap Map works as a radar. A team weighing whether to base its product on an open model or library can see what alternatives exist at each layer and which are barely covered. For foundations and funders, the map points to where resources should go: the gaps are, literally, the shopping list. And for researchers, it offers a picture of the real split between software, models, datasets and hardware, four blocks that are rarely looked at together.

It is worth being realistic about the scope. This is a version 0.1, with 421 projects reviewed out of nearly 25,000 detected. Its usefulness will depend on Current AI keeping up the cataloguing pace and on the community contributing, because an outdated gap map misleads more than it guides.

At ElephantPink we work daily with MCP servers, models and open libraries, so we welcome any serious attempt to bring order to an ecosystem that grows faster than anyone can follow. The Gap Map strikes us as a sensible starting point, above all for what it admits it does not yet know. We will have to see whether in six months it is still alive or joins the long list of directories that were born with good intentions and froze at version 0.1.