Free LLM API tiers change without warning — a 30-line stdlib script to check which ones still work

Free tiers on LLM APIs are a moving target. A provider that worked last week silently drops to a paid-only plan, rate-limits harder, or deprecates the model you pinned — and the first sign is usually a production failure, not a changelog entry.

Instead of finding out from a stack trace, ping every configured endpoint on a schedule and check three things: HTTP status, latency, and whether the response actually contains content.

import json, time, urllib.request

def check_endpoint(name, base_url, api_key, model):
    t0 = time.time()
    req = urllib.request.Request(
        f"{base_url}/chat/completions",
        data=json.dumps({
            "model": model,
            "messages": [{"role": "user", "content": "ping"}],
            "max_tokens": 5,
        }).encode(),
        headers={"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"},
    )
    try:
        with urllib.request.urlopen(req, timeout=15) as r:
            data = json.loads(r.read())
            latency = round(time.time() - t0, 2)
            choice = (data.get("choices") or [{}])[0]
            content = choice.get("message", {}).get("content", "")
            finish = choice.get("finish_reason")
            ok = bool(content.strip())
            return {"name": name, "ok": ok, "status": r.status, "latency_s": latency, "finish_reason": finish}
    except Exception as e:
        return {"name": name, "ok": False, "error": str(e)}

The trap worth calling out explicitly: a 200 OK with an empty content and finish_reason: "length". That's not a failure by HTTP status — it's a reasoning model that burned its whole token budget on hidden reasoning tokens and never emitted an answer. If you only check status_code == 200, this one passes your health check and then fails in production.

Run this against every provider in your pool once a day (or before each deploy) and log the result — ok, latency_s, finish_reason. Three runs of silent failure is your signal to fail over, not the 2am page.

This stdlib script is the minimum viable version. If you want a maintained, curated list of free-tier LLM providers — which ones are currently genuinely free with no card, their rate limits, and known gotchas like this one — I packaged that research as the Free-LLM Provider Pack.