Get the weekly radar
One short email with the freshest products, every week.
Grounded in available product and source data
inferock-bench is a local cost-tracking proxy for LLM API calls to OpenAI, Anthropic, Gemini, and pinned OpenRouter models. Each call gets an independent, per-call receipt of what was billed and what failed, so developers can audit an AI or LLM bill call by call.
The core use case is measuring Claude or GPT token usage locally and checking whether a failed API call still got billed, giving developers visibility into their LLM spend independent of the provider's own reporting.
It cannot cap provider spend on calls it never sees, and it can't explain traffic that bypassed the local proxy — so it's a record of what passed through, not a hard spending limit.
inferock-bench runs locally via `npx inferock-bench`, sitting in front of your existing provider keys. Those keys stay on your machine and are only ever attached to outgoing provider requests — they're never sent to Inferock itself.
It's useful for spotting a request that timed out but still shows usage, or token counts that don't line up with the visible output — patterns that are easy to miss if you're only looking at a provider's summary totals.
It can help trace retries that may have quietly driven up cost, along with latency or model-version changes that leave a trail worth checking rather than assuming everything ran as expected.
No — it pins specific endpoints such as meta-llama, deepseek, mistral, moonshot/kimi, z-ai/glm, and qwen on observed hosts, rather than covering the entire OpenRouter catalog.
