The Open Source AI Landscape: How the Gap Map Charts the Ecosystem

A New Map of the Open Source AI Landscape

The open source AI ecosystem is expanding at a breathtaking pace, yet it has long lacked a systematic, panoramic view — which models, tools, datasets, and hardware projects truly form the backbone of this ecosystem? The Open Source AI Gap Map, recently released by nonprofit organization Current AI, sets out to answer exactly that question.

Current AI describes itself as "a global partnership building a public option for AI." The concept of a "Public Option for AI" is borrowed from public policy: it refers to AI infrastructure led by nonprofits or governments, not primarily driven by commercial interests — analogous to public broadcasting, public libraries, or open source operating systems. The idea is to provide a publicly accessible, auditable, and usable alternative in a market dominated by private giants. Current AI was formally established as a nonprofit at the Paris AI Action Summit in February 2025 — one of the most significant multilateral AI governance events since the 2023 Bletchley Park Summit in the UK, and a platform the French government positioned as a key driver of international cooperation on "trustworthy AI." With pledged funding of up to $400 million, the scale of backing — from governments, foundations, and research institutions — signals this is far more than a grassroots community project. With that kind of capital and pedigree, the goal is clearly not just to build an index, but to establish trusted public infrastructure for open source AI.

Core Data in Gap Map v0.1

According to official figures, Gap Map v0.1 provides in-depth coverage of 421 products, distributed as follows:

• 266 software tools and code libraries
• 85 models
• 50 datasets
• 20 hardware projects

These products are produced by 228 organizations and are organized into 14 categories spanning three layers of the technology stack (model components, products/UX, and infrastructure).

One important detail: those 421 are just the tip of the iceberg — the deeply researched items. The project actually tracks approximately 24,800 entries in total, with the remaining ~24,400 forming the long tail of the open source AI ecosystem, entries that won't be scored until they've been researched and reviewed. That number alone illustrates just how vast and fragmented the open source AI landscape truly is.

Why It's Called a "Gap Map"

The name "Gap Map" carries real methodological weight. It draws on the practice of Evidence Gap Maps from social science and policy research, systematically developed by international development evaluation bodies such as the UK's DFID and the Campbell Collaboration. The core logic is this: within a given intervention domain, systematic reviews identify areas where evidence is robust and areas where it is absent — guiding resource allocation and research prioritization.

Applying this methodology to AI ecosystem analysis means the project is not merely a catalog of "what exists," but an explicit inquiry into "what's missing." By mapping the full technology stack in layers, it allows anyone to see at a glance which parts have mature open source solutions and which remain blank or underserved. The Gap Map presupposes an ideal "complete technology stack" and then marks which segments still lack mature open source alternatives — giving funders, researchers, and policymakers an actionable basis for setting priorities.

This perspective is especially valuable for policymakers, researchers, and enterprise decision-makers alike. When people debate whether AI should have a public option, or whether open source can truly compete with closed-source giants, they rarely have a quantifiable, traceable evidence base to draw on. The Gap Map is precisely an attempt to fill that void — translating vague industry intuitions into structured data.

The Three-Layer Stack Classification

Organizing products into three layers — "model components," "products/UX," and "infrastructure" — is a pragmatic and useful framework:

• Model components: Foundation models, datasets, and other elements that constitute core AI capabilities
• Products/UX: Application-layer tools facing end users
• Infrastructure: The underlying facilities and hardware that support training, deployment, and inference

This layered approach helps illuminate the full open source AI value chain. A healthy open source ecosystem can't consist only of popular models — it requires complete data pipelines, toolchains, and infrastructure to support them.

The Real Value: Open Underlying Data

The map itself is well worth exploring, but as veteran developer Simon Willison has pointed out, the truly exciting part is the underlying data.

Current AI has open-sourced all the data under the MIT license, published in the currentai-org/os-ai-map GitHub repository, which includes:

• 1,184 YAML files (structured product entries)
• Notebooks used for data collection
• A data schema
• Various helper scripts

The choice of YAML (YAML Ain't Markup Language) as the storage format reflects deep-rooted preferences in open source engineering culture. YAML is prized for its human readability — it's far more suited to manual editing and Git-based version control than JSON, with each change producing a clean diff that the community can easily review and contribute to. More critically, the design of 1,184 individual YAML files (one per product) rather than a single database means any contributor can submit new entries or correct existing ones via a standard Pull Request workflow. This "files as database" collaboration pattern is widely used by mature open source projects like Homebrew and the awesome-* repositories — a convention that has been refined through years of community practice.

This means anyone can reuse, audit, extend, or even challenge this data. For a project committed to "transparency" and "public good," open data is itself the most powerful statement of intent. Compared to leaderboards that offer polished front-ends while locking away the underlying data, a complete dataset under the MIT license genuinely grants the community the right to participate.

Instant Exploration with Datasette Lite

Because the files are hosted on GitHub, anyone can explore them directly in a browser using Datasette Lite — no local installation required. Datasette Lite is a browser-based SQLite data exploration tool built by Simon Willison (co-creator of Django) using WebAssembly technology. Its core stack leverages Pyodide (a WebAssembly port of Python) and sqlite-wasm, allowing a full SQLite database engine to run directly inside the browser sandbox with no server-side computation. This approach transforms data exploration from "requires a deployed service" into "share a URL" — dramatically lowering the distribution cost of data analysis.

Simon Willison demonstrated one typical use case: the project's 16,185 tracked GitHub repositories are loaded as a CSV file that can be immediately sorted, filtered, and queried by star count. This zero-friction exploration workflow is exactly what makes open data compelling — once data is released, the surrounding tool ecosystem can immediately empower it, allowing anyone who's curious to run their own analysis, rather than depending on a single official conclusion.

Significance and Outlook

Gap Map is still at v0.1, with the vast majority of entries yet to be scored and deeply researched — the long tail remains largely uncharted territory. But it has made a strong start:

1. It provides a public, auditable evaluation framework, giving discussions about open source AI a concrete evidence base;
2. It practices openness through open data, avoiding the contradiction of "advocating for open source while keeping data closed";
3. It lowers the barrier to ecosystem exploration, enabling ordinary developers to participate in analysis using tools like Datasette Lite.

For anyone tracking the trajectory of open source AI, this map is worth following closely. As more entries are researched, scored, and cited, it has the potential to become an important reference for measuring the health of the open source AI ecosystem — a continuously updated "ecosystem health report." At a time when closed-source large models increasingly dominate the discourse, this kind of commitment to public interest and transparency is especially valuable.

Key Takeaways