Discover
Poll registered official APIs and changelogs. Community links can suggest a candidate but cannot verify one.
“Found on the web” and “working free API” are different claims. FreeToken keeps discovery, two-provider review, verification and publication separate.
Machine-readable catalogs run more often; slower documentation pages use a longer interval to reduce noise and cost.
| Source | Provider | Signal | Trust | Planned cadence |
|---|---|---|---|---|
| OpenRouter Models API | openrouter | model catalog | Official evidence | 1 h |
| OpenRouter API changelog | openrouter | changelog | Official evidence | 3 h |
| Gemini API release notes | google-gemini | changelog | Official evidence | 3 h |
| Groq supported models | groq | model catalog | Official evidence | 3 h |
| Groq API changelog | groq | changelog | Official evidence | 3 h |
| NVIDIA API Catalog | nvidia-nim | model catalog | Official evidence | 6 h |
| NVIDIA NIM LLM release notes | nvidia-nim | release notes | Official evidence | 6 h |
| Cerebras Inference changelog | cerebras | changelog | Official evidence | 3 h |
| SiliconFlow release notes | siliconflow | release notes | Official evidence | 6 h |
| 2nothing AI agent discovery document | 2nothing | developer docs | Official evidence | 12 h |
| awesome-freellm-apis community index | multi-provider | discovery index | Discovery only | 6 h |
10 provider-owned sources can support publication. 1 community index can surface new leads, but it is deterministically blocked from auto-publishing until an official source corroborates it.
Poll registered official APIs and changelogs. Community links can suggest a candidate but cannot verify one.
Normalize each source, hash the meaningful content and create a candidate only when it changed.
A free model extracts a strict evidence object only when the meaningful source hash changes.
A platform-owned credential tests the exact model route without using a visitor's API key.
A different provider independently checks the source, conditions, free-access claim and Canary result.
Deterministic risk, confidence and topic-fit rules approve, hold or reject the candidate.
High-value official changes become original practical analysis; smaller facts become evidence updates; noise is ignored.
Update the catalog, model page, Blog/News, RSS and Watchlist from one approved event.