Adil Islam
📡 Signal Analysis

The Open Source AI Gap Map

Current AI mapped 24,626 open source projects across the entire AI stack. The map reveals where open source is thriving — and where critical gaps remain.

July 4, 2026 · Based on Simon Willison's coverage of Current AI's Gap Map v0.1
24,626Projects Surveyed
421Deep-Scored Products
228Organizations
$400MCommitted Capital

Current AI — a non-profit founded at the AI Action Summit in Paris in February 2025 — launched the Open Source AI Gap Map v0.1 on July 1. Backed by $400M in committed capital, the project aims to make the sprawling open source AI ecosystem legible: where is it strong, where is it fragile, and — most importantly — where should builders and funders focus next.

What the Map Shows

The Gap Map organizes the open source AI stack into 14 categories across 3 layers: model components, product/UX, and infrastructure. Of the 24,626 artifacts identified, 421 were deeply scored across three axes:

The most interesting findings live where these three axes disagree — the widely-used model that's barely open, or the fully-open project nobody uses.

🔍 Early Findings

Open source isn't chasing the frontier — it's ahead of it. Entire capability categories like orchestration agents were first developed in the open source ecosystem, not by frontier labs. The open source AI economy is out-innovating the closed one.
Contributors are building shared infrastructure, not free riding. This isn't a tragedy of the commons — contributors are actively investing in shared tooling, signalling ecosystem health that's easy to underestimate from the outside.
Health and resilience aren't the same thing. Inference is a case study: vLLM, llama.cpp, and SGLang are mature, well-adopted, and genuinely open. But there are only a handful of them. The entire inference layer runs on a bus factor of a few key projects. This structural vulnerability repeats across the stack.

What Makes This Different

The Gap Map deliberately avoids pitting open vs. closed AI ecosystems against each other. Instead, it maps what's needed to build the system the community wants — a constructive framing that shifts the conversation from "we're behind" to "here's exactly where to invest."

The taxonomy descends from the 2024 Columbia Convening on Openness in AI, giving it academic rigour grounded in community consensus.

Simon Willison's Take

Simon, who linked to the map on July 3, highlighted what excites him most: the underlying data. The entire dataset — 1,184 YAML files, notebooks, schemas, and collection scripts — is released under an MIT license at currentai-org/os-ai-map on GitHub.

He immediately pointed readers to a Datasette Lite instance exploring the 16,185 GitHub repos the project is tracking — sortable by stars, browsable by any dimension. This is classic Willison: the map is interesting, but the real unlock is giving the community raw, queryable data to find their own answers.

Why This Matters Now

The open source AI ecosystem has reached a stage of maturity where it's both seriously capable and structurally fragile at the same time. Models like GLM-5.2 and DeepSeek V4 are competitive with frontier closed models. Inference engines serve production workloads for major companies. But as the Gap Map makes explicit, that capability is concentrated in a dangerously small number of projects.

For builders, the message is clear: there are well-defined gaps — in observability, evaluation, orchestration, training tooling, and more — where a well-executed open source project could have outsized impact. Current AI is inviting collaborators, editors, and contributors to shape future iterations.

→ Explore the map at map.currentai.org
→ Read the full announcement: Current AI Blog
→ Dive into the data: GitHub Repo
→ Simon's original post: Open Source AI Gap Map

Signal by Adil Islam · Daily AI research curation · Browse archive