Inferock-bench is a local proxy that sits between your app and OpenAI, Anthropic, Gemini, or OpenRouter shaped calls. It captures per-call token usage, failures, and retries, then generates an independent receipt showing what you were billed and how much you're actually overpaying for.
This is the 2nd launch from Inferock. View more
Inferock
Launched this week
Get free signup credits to use all available models on Inferock.
Bring your own keys: run your OpenAI, Claude or Gemini accounts, see where calls fail and what they cost, and get credits back on eligible failed calls.
Serverless: pay-per-token LLM inference. Get signup credits to start using.
Free Options
Launch Team / Built With
Love the focus on transparency and accountability. Wishing the team great success! 👏
@chrishicken Thank you, that means a lot.
Accountability for production AI is exactly what we set out to build. If you try Inferock Watch on your own keys, we would love to hear what you think!
what counts as an "eligible" failed call? our worst week was a provider handing back 429s under load while our own retry backoff quietly burned the whole time budget, so the user saw a fallback and we still paid for the attempts that did go through late. would a 429, or a call we cancel ourselves on timeout, get credit back? and does the receipt show retries per request or just totals?
Congrats Bharath, Hamza, and the Inferock team on the launch. The Watch-as-local-proxy model with your own keys is the part that makes receipts usable without piping traffic through another dashboard. When a call returns valid JSON but still fails your app's required field check, does that show up as the same class of credit-eligible failure as a timeout, or is "completed but unusable" tracked separately on the receipt?
@genedai Thank you. Today the receipt tracks only what the gateway can see: errors, timeouts, and empty answers that billed tokens. Those are credited automatically. Valid JSON that fails your own field check looks complete to us, so it is not credited or marked separately yet. Marking a call unusable from your side is a sensible next step, and we'll try to acomodate this pattern in the future!
Love the idea of an independent receipt for LLM calls, getting credits back on failed calls is a refreshingly user-first take on inference billing.
@ilko_kacharov Thank you, that was exactly the idea.
To be precise about the credits: they apply to eligible failures, they are non cash credit, and they work within the balance, limits and expiry in our terms.
On Inferock Watch they come through a separately enrolled program, and on Inferock Serve under agreed terms.
The details are at inferock.ai/pricing, and we are happy to answer questions here.
@ilko_kacharov Thank you! Hope you give us more feedback in the coming days!
I like that credits only cover objectively broken calls timeouts, truncated output, bad JSON, not just “the model gave a bad answer.”
@hamza_aafzal Thank you, that line mattered to us: credit should follow failures anyone can verify.
Eligible failures earn non cash credit within the balance, limits and expiry in our terms, on Inferock Watch through a separately enrolled program and on Inferock Serve under agreed terms.
The exact list is at inferock.ai/pricing.