Skip to content

7.4 billion tokens per month. 34 free LLM providers. 635 free model endpoints. All behind one /v1 endpoint, plus any custom OpenAI-compatible endpoint. Smart routing, automatic failover, encrypted keys. Personal experimentation only.

TypeScript
2.6 KB
Raw Blame

Router9 and Septor Labs

Both are first-class OpenAI-compatible chat providers. Select them on the Keys page or import ROUTER9_API_KEY / SEPTOR_API_KEY. This integration does not seed model rows: the signed Oracle catalog controls availability and the existing Premium-immediate / Free-after-30-days release gate.

Free access (checked 2026-09-10)

  • Router9: 50,000 shared credits (about $5) each month on Free. Only provider-allowlisted models work; the tested route is minimax/minimax-m3. auto currently returns 404. Paid top-ups are optional and outside the free allowance.
  • Septor Labs: zero-price free models share 50 requests / 500K tokens daily. Its separate $1 signup grant expires after 90 days and is not a recurring monthly grant. Tested matching routes: hermes-3-llama-3.1-405b-free, minimax-m2.5-free, qwen3-coder-free, and qwen3-next-80b-a3b-instruct-free.

Integration details

Router9's /v1/models is public even with an invalid bearer. Key validation instead sends an incomplete, no-generation request to /v1/chat/completions. A valid key yields 404 model_not_found with account credit headers; an invalid key yields 401. Validation consumes request quota, not inference credits. Unknown verdicts (including generic proxy 404/429) remain inconclusive rather than disabling a valid key. Streaming reuses the existing think-tag filter and EOF handling; Router9 sends a finish reason and usage but no [DONE] sentinel. Its live request-limit headers disagree with its documentation, so no fixed request limit is invented. Actual credit headers feed the shared quota pool.

Septor authenticates /v1/models. Eleven other free IDs returned MiniMax M2.5 instead of the requested model during testing. Its adapter rejects mismatched or missing response model IDs with a retryable 502, including before yielding stream content, instead of silently reporting a substituted model. Explicit auto routing may report a different model; it is not seeded or published by this change. Model-ID matching is provider response evidence, not independent verification of underlying weights.

No keys, paid plans, catalog snapshots, DB model migrations, or release-gate changes are included in this integration.

The compiled adapters passed ten live inference checks (normal and streaming for all five routes), plus valid/invalid key checks for both providers. Septor reported zero charged cost for all eight adapter inference checks. All ten answers completed with the expected text; these are smoke tests, not benchmarks.