I've been running hourly HTTP checks against 99 free public APIs for the past
few months. The dataset started as a weekend project and turned into an
uncomfortable audit of the most popular API resource on GitHub.
The problem with the big list
The public-apis/public-apis repo
(350k+ stars) is the canonical "find a free API" resource. It's also a markdown
file — nothing verifies that yesterday's entries still work today.
Six APIs that are prominently findable through that list and its downstream
mirrors are dead or paid in 2026:
| API | What happened | Still listed? |
|---|---|---|
| Dark Sky | Killed Dec 2022 (Apple acquisition) | mirrors: yes |
| Twitter v1.1 free | Paywalled July 2023 ($100/mo min) | mirrors: yes |
| Reddit free API | Effectively killed June 2023 | mirrors: yes |
| Yahoo Finance v7 | Dead since 2017, unofficial wrappers blocked | yes |
| MetaWeather | Offline since April 2022 | mirrors: yes |
| Weather Underground | Free tier sunset 2019 (IBM) | mirrors: yes |
From my monitoring data, conservatively 15-25% of entries on any large
community API list are degraded — dead, paid, moved endpoints, or
rate-limited far below what the docs claim.
Why this happens (it's structural, not negligence)
A markdown row, once merged, persists forever unless someone files a removal
PR. Removal PRs are unrewarding — they don't grow the catalog or generate
stars — so they rarely happen. Meanwhile the economics of "free" APIs keep
producing casualties: startup launches generous free tier → tier becomes a
real AWS line item → monetization or acquisition kills it.
The list isn't bad. Discovery and verification are just different jobs.
What verification looks like
My setup (runs entirely on Cloudflare's free tier, $0/month):
- A Worker on two cron schedules pings each API's documented health endpoint hourly (two crons because the free tier caps you at 50 subrequests per invocation — 99 APIs don't fit in one)
- Results go to D1; a builder Worker exposes a JSON snapshot
- An Astro static site rebuilds daily with 30-day uptime per API
Fun findings from the data:
- Most reliable: government-backed and community infrastructure. MET Norway (Norwegian met service), Open-Meteo, ipify — 99.9%+.
- Least reliable: httpbin.org (absorbs the entire internet's CI traffic), and venture-funded free tiers generally.
- Weirdest failure mode: APIs that return 200 with an HTML error page, which naive uptime checks count as "up."
The workflow that actually works
- Discover on the GitHub list (breadth is unbeatable)
- Verify before integrating: curl the endpoint with your exact request shape, check the docs URL still resolves, look for a status page
- Monitor after shipping: any uptime tool pointed at the endpoints you depend on
I publish my monitoring data at freeapi.watch —
99 APIs, hourly checks, plus a graveyard
of the dead ones with replacements. The JSON is public if you want to build
on it.
What's the deadest "free" API you've been burned by? The graveyard takes
submissions.
Top comments (0)