Inferock Bench

Your AI bill can be quietly wrong.

  Providers shouldn't get to grade their own bills.
  Today, the company that charges you also decides what counts as a failure, what gets credited, and keeps the only detailed records. inferock-bench puts an independent, per-call receipt of what you were billed — and what failed — in your own hands.

  Receipt contract ·
  Quickstart ·
  Integrations ·
  Test your loss ·
  What can go wrong ·
  How this compares ·
  Key boundary ·
  Docs
Disclosure: Inferock sells a hosted product, and this benchmark grades receipts against The Inferock Standard, which Inferock wrote. Do not take that on authority: check the public run cards in docs/ (the underlying sanitized event archives are not yet published), the spec, and the source, and read the hard questions and threat model for limits and adversarial-review invitations.

Illustrative example, not a measured run. This lead image shows what a $124,863.47/mo AI spend could surface for review when a defensibly calibrated synthetic scenario is passed through the real bench pipeline. The pipeline organically aggregates $5,708.25 in money loss, including $4,564.14 estimated recoverable and $1,144.11 not recognized yet. Not measured by Inferock, not a guarantee, not a bill audit.

Basis: independent third-party billing-audit inputs plus public latency throughput/SLA thresholds; source links and caveats are one click away.

See the basis -> · See newest measured run ->

Use it when you need to audit an AI/LLM bill, measure Claude or GPT token usage locally, or answer "was I billed for a failed API call?" It is a local LLM cost-tracking proxy for four measured provider planes: OpenAI, Anthropic, Gemini Developer API, and pinned OpenRouter endpoints spanning meta-llama, deepseek, mistral,…