Current AI evaluated 24,626 open source AI projects, from foundation models through inference backends, and published the results as the Gap Map, a public tool at map.currentai.org for seeing the open source AI stack as one system rather than a scattered list of repositories. Each project is assessed across openness, capability, and adoption. The project draws on prior work from the Columbia Convening, MOF, and Hugging Face, and Current AI describes the underlying stack as robust but fragmented and hard to see whole.
That framing names a real cost. Fragmentation in open source infrastructure does not just mean duplicated code. It means duplicated funding decisions, duplicated maintainer burnout, and duplicated due diligence, because every team evaluating an inference backend or an eval framework has to rediscover which projects are alive, which are abandoned, and which quietly do the same job as three others. Without a shared reference, a foundation deciding where to put a grant and a startup deciding which library to depend on run the same research independently, often reaching different, uncoordinated answers.
A map changes that by turning a private judgment call into a shared one. When 24,626 projects are scored on the same three axes, a funder can see not just what exists in a category but where capability is thin relative to demand, and a builder can see whether a gap is truly empty or merely under-documented. That is the difference between “nobody has built this” and “several people built this badly and none of them talk to each other,” and only a systematic assessment across the full stack can tell the two apart.
The Gap Map’s value depends entirely on adoption as a coordination reference rather than a one-time report. Current AI is explicitly recruiting collaborators to review products, refine the methodology, and add tools, which signals the project understands that a snapshot of 24,626 projects goes stale the moment new models, backends, and eval harnesses ship. A map that is not maintained collectively will drift out of date faster than any single team can track, and its usefulness as a coordination artifact collapses the day it stops reflecting the current stack.
The open source AI stack does not lack activity. It lacks a shared view of where that activity concentrates and where it thins out, which is precisely the gap between individual project momentum and collective strategic direction. For a founder or program officer weighing where to place the next grant, contribution, or engineering hour, the Gap Map is a reason to check where a category scores low on adoption or openness before assuming someone else has already covered it, and a reason to submit missing tools rather than build in isolation from what the map already shows.
Current AI’s Gap Map at map.currentai.org, published in 2026.
