Verifiable, not asserted

Model Provenance & Billing Integrity

The #1 fear when buying from an AI API gateway is a silently downgraded model or a hidden markup. Here is exactly how NovAI removes both — and how you can verify it yourself.

Official-direct upstream, zero markup

0% platform feeofficial-direct

Every model is served from the provider’s official API at the provider’s list price. NovAI adds no markup (vs OpenRouter ~5.5%). The price you see on the leaderboard is the price you pay.

Model IDs are verifiable, not marketing

The live model list (121 models right now) is served at GET /v1/models. Each id matches the provider’s official model id, so you can cross-check any model against the provider’s own docs. There is no anonymous “turbo/mini” relabelling.

curl https://aiapi-pro.com/v1/models

No silent downgrade — and it costs you nothing if anything fails

Requests route to the exact model id you specify, and the response reports the real upstream usage. Video and 3D jobs that fail are refunded in full, and unused balance is refundable within 7 days — so neither failure nor mismatch is ever your cost.

Billing you can recompute yourself

Charge = upstream usage.total_tokens / 1000 × published token price. There are no hidden cache or platform charges; cache-hit billing follows exactly what the upstream reports in usage.

  1. Call a model and read response.model and response.usage.total_tokens.
  2. Multiply by the published per-1M price on the leaderboard.
  3. Compare to the credits deducted in your dashboard — they match.

Independently measured & third-party checked

Machine-readable: /provenance.json · generated 2026-10-11 19:20 UTC · related: transparency, status, leaderboard.