Free LLM APIs, actually verified.
Every “free LLM API” list goes stale the week it’s published. ModelHubby is a maintained registry: normalized records with real limits, signup friction, and data policies — every fact carrying provenance, checked by live probes, and versioned so you can see what changed.
Not another list
The registry treats provider facts like infrastructure, because your fallback chain depends on them.
Verified, not scraped
Three-level checks — record sanity, live model listings, and tiny real completions. When Google’s model list advertised a model that 404’d on chat, the probe caught it the same day.
“Free” ≠ frictionless
Card, phone, membership, time-to-first-key, and prompt-training policies are first-class fields. Filters never treat “unknown” as a pass.
Quota units, uncoerced
RPM, RPD, tokens, credits, neurons/day, concurrency — kept distinct, with the provider’s raw wording preserved next to parsed values.
Routing-ready
A local OpenAI-compatible gateway routes model: “auto” across your keys with health-weighted scoring, per-key cooldowns, and circuit breakers.
Provenance on every fact
Records cite their sources with trust tiers and check timestamps. Conflicts between sources are preserved and flagged, not silently resolved.
Versioned churn tracking
Monthly snapshots and material-change reports: which providers tightened limits, which models vanished, which policies flipped.
Featured free tiers
Ranked views, full records, and models for all 42 in the directory.