The editorial argues that the gap between curated lists and actual software is widening precisely because LLMs hallucinate APIs, cite deprecated endpoints, and confidently return 404s. Human-maintained directories like public-apis and free-programming-books act as a trusted ground-truth layer that developers route through before asking a model to write integration code.
freeCodeCamp's 437.9k stars reflect that its real product is the curriculum — an organized index of exercises, articles, and certifications — not the JavaScript platform behind it. The implicit position: a structured learning path applies to every developer regardless of stack, which is why it sits at #1.
The repo is literally a markdown file of free APIs, and its 411.9k stars suggest developers value a maintained, PR-vetted index over search engines or AI chatbots. The position: a list curated by humans is more trustworthy than any generated answer when the cost of a wrong link is wasted integration time.
free-programming-books reached 384.0k stars as a pure markdown directory of free learning resources. The position: curation by topic and language outperforms search because someone has already filtered, organized, and kept it current.
awesome-python at 286.7k stars is the fourth-most-starred repo and a textbook curated list — an opinionated index of Python frameworks and libraries. Its position just below the top three reinforces the pattern: lists dominate, and language-specific lists are the next tier.
281.2k stars for a list of self-hostable services shows the awesome-list format also wins in operational/devops niches. The implicit position: when developers want to find software in a category, they trust a community-curated markdown file over a search query.
awesome-go at 167.1k stars repeats the pattern for the Go ecosystem. The position: every major language develops a flagship curated list that becomes the de facto discovery layer for its community.
free-for-dev at 120.6k stars curates SaaS/PaaS/IaaS free tiers — another markdown index that vastly outranks most actual software. Reinforces the position that lists win even in commercial-tooling discovery.
developer-roadmap at 350.5k stars sits in the same tier as the top three by being a visual index of what to learn and in what order. The position: developers reward content that makes the chaos of the ecosystem navigable, whether it's a markdown list or an interactive roadmap.
HelloGitHub at 147.6k stars curates entry-level open source projects for a Chinese-speaking audience. The position: discovery and onboarding content is as valuable as the software being discovered, and crosses language and cultural boundaries.
React at 243.9k stars is the highest-ranked actual framework, but still trails the top three lists by ~150k stars. The implicit position: even ubiquitous software has a narrower audience than universal indexes — you only star React if you write JS.
The Linux kernel at 221.6k stars is arguably the most-used software on Earth, yet still ranks below several markdown files. The position: stars measure aspirational bookmarking, not usage — and a list is something everyone might want, while a kernel is something only some will touch.
VS Code at 182.5k stars is the most-used IDE on the planet but still ranks well below curated lists. Reinforces the position that real software has a usage ceiling tied to its niche, while lists transcend niche.
The top of GitHub's star leaderboard hasn't moved in years, and the composition tells a quieter story than the numbers. As of this week, the three most-starred public repositories are freeCodeCamp/freeCodeCamp (437.9k★, a learning curriculum), public-apis/public-apis (411.9k★, a list of free APIs), and EbookFoundation/free-programming-books (384.0k★, a list of free books). Combined: 1.23 million stars.
None of them ship a runtime. None of them are a framework, a language, a database, or a build tool. All three are curated indexes — human-maintained directories of links, organized by topic, kept current by pull request. freeCodeCamp has a JavaScript codebase behind its platform, but the thing people star is the curriculum: a list of exercises, articles, and certifications. The other two are markdown files. That's the whole product.
This is not a 2026 phenomenon. These three have occupied the top of the leaderboard, in roughly this order, for most of the last five years. What's new is the gap. The fourth-place repo, sensiml/awesome-python, sits at 286.7k. The drop-off from #3 to #4 is nearly 100k stars. The lists aren't just winning — they're pulling away.
The naive read is that stars are a popularity contest and curated lists win popularity contests because they're useful to everyone. That's true and uninteresting. The interesting question is why curation is *accelerating* its lead in a year when generative AI was supposed to make indexes obsolete. If you can ask Claude or GPT for "a free public weather API," why is public-apis still adding ~30k stars a year?
Because LLMs hallucinate APIs that don't exist, return endpoints that 404, and confidently cite deprecated services with the same tone they cite live ones. A junior developer who trusts the model loses an hour. A senior one who's been burned twice routes the query through a maintained list first, then asks the model to write the integration. public-apis isn't competing with ChatGPT — it's the trusted lookup table ChatGPT should have been grounded against in the first place. Every PR that removes a dead endpoint is a small RAG correction the LLM ecosystem can't perform on itself.
The same logic applies to free-programming-books. Anthropic and OpenAI have ingested every PDF in that repo a hundred times over. The repo's value isn't the books — it's the *human judgment about which books are worth reading*, organized by language and updated as the field moves. An LLM can summarize a book it's already trained on; it cannot tell you which of 400 Rust books is the one a working engineer in 2026 should actually open first. That's an editorial decision, and editorial decisions are exactly what training data flattens.
freeCodeCamp is the outlier of the three — it does have a real platform — but the star count is for the curriculum, not the React app. Quincy Larson's team makes the call about what a self-taught developer should learn, in what order, with what exercises. That sequencing is the moat. AI tutors can answer any question; they can't tell you which question to ask next on day 47 of a career change. Curation is sequencing, and sequencing is taste.
The community reaction to this stratification has been muted, which is itself the signal. There's no discourse about it because no one disputes it. The most useful artifacts on GitHub are the ones with the lowest technical complexity and the highest human-judgment density. That's a strange shape for a code-hosting platform.
Three concrete implications.
First, if you're shipping a developer tool in 2026, the gatekeeper isn't a benchmark or a Show HN — it's whether you land on the relevant awesome-list. Inclusion in a curated index is now worth more downstream traffic than a 500-comment HN front page, because the list compounds and the HN thread doesn't. Audit the awesome-* repos in your category, open the PR yourself, write the one-line description as if a maintainer is going to merge it without edits (because they will or won't, on that basis alone).
Second, if you're building anything that depends on third-party APIs, treat public-apis as a first-pass lookup before you trust an LLM's suggestion. The repo has 6,500+ entries, all with at least one human PR review, all rechecked when issues are filed. It's slower than asking Claude, and correct more often. The combination — list first, model second — is the workflow that actually works.
Third, if you maintain documentation, ship it as a curated index, not as auto-generated reference. Auto-generated docs are commodity content the moment a model ingests them; an opinionated "read these five pages in this order" is a moat the model can't replicate without your judgment. freeCodeCamp's curriculum is the proof. Sequencing is the product.
The leaderboard probably won't change shape this decade. As more code becomes machine-written, the human-curated map of where the useful things are will get more valuable, not less — and the stars will continue to flow to the maps. The lesson for anyone building in public is uncomfortable: your best distribution channel may not be your repo. It's the list your repo gets added to.
A collective list of free APIs
→ read on GitHubfreeCodeCamp.org's open-source codebase and curriculum. Learn math, programming, and computer science for free.
→ read on GitHub:books: Freely available programming books
→ read on GitHubYour own personal AI assistant. Any OS. Any Platform. The lobster way. 🦞
→ read on GitHubInteractive roadmaps, guides and other educational content to help developers grow in their careers.
→ read on GitHubA list of Free Software network services and web applications which can be hosted on your own servers
→ read on GitHubAn opinionated list of awesome Python frameworks, libraries, software and resources.
→ read on GitHubSkills for Real Engineers. Straight from my .claude directory.
→ read on GitHubThe agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
→ read on GitHubThe agent that grows with you
→ read on GitHubLinux kernel source tree
→ read on GitHubThe library for web and native user interfaces.
→ read on GitHubThe open source coding agent.
→ read on GitHubFair-code workflow automation platform with native AI capabilities. Combine visual building with custom code, self-host or cloud, 400+ integrations.
→ read on GitHubA feature-rich command-line audio/video downloader
→ read on GitHub🔥 The API to search, scrape, and interact with the web for AI
→ read on GitHubAn Open Source Machine Learning Framework for Everyone
→ read on GitHubVisual Studio Code
→ read on GitHubA curated list of awesome Go frameworks, libraries and software
→ read on GitHub🙃 A delightful community-driven (with 2,400+ contributors) framework for managing your zsh configuration. Includes 300+ optional plugins (rails, git, macOS, hub, docker, homebrew, node, php, python
→ read on GitHubAutoGPT is the vision of accessible AI for everyone, to use and to build on. Our mission is to provide the tools, so that you can focus on what matters.
→ read on GitHubGet up and running with Kimi-K2.5, GLM-5, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.
→ read on GitHub:octocat: 分享 GitHub 上有趣、入门级的开源项目。Share interesting, entry-level open source projects on GitHub.
→ read on GitHubFlutter makes it easy and fast to build beautiful apps for mobile and beyond
→ read on GitHubf.k.a. Awesome ChatGPT Prompts. Share, discover, and collect prompts from the community. Free and open source — self-host for your organization with complete privacy.
→ read on GitHubThe most popular HTML, CSS, and JavaScript framework for developing responsive, mobile first projects on the web.
→ read on GitHub🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training.
→ read on GitHubProduction-ready platform for agentic workflow development.
→ read on GitHubA modern GUI client based on Tauri, designed to run in Windows, macOS and Linux for tailored proxy experience
→ read on GitHubUser-friendly AI Interface (Supports Ollama, OpenAI API, ...)
→ read on GitHubLangflow is a powerful tool for building and deploying AI-powered agents and workflows.
→ read on GitHubThe agent engineering platform
→ read on GitHubCollection of publicly available IPTV channels from all over the world
→ read on GitHubThe React Framework
→ read on GitHubMicrosoft PowerToys is a collection of utilities that supercharge productivity and customization on Windows
→ read on GitHubA list of SaaS, PaaS and IaaS offerings that have free tiers of interest to devops and infradev
→ read on GitHubTop 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.