Reading your NeverMetered usage and requests pages

Updated 10 November 2026

You're never billed per token on NeverMetered, so the Usage and Requests pages aren't a bill. They're there so you can see what your agents are doing, why something felt slow, and whether you need more parallel streams. Every number below comes from the pages as they appear in your dashboard.

The Usage page

The Usage page with eight tiles: input tokens, cache hit rate, output tokens, requests, GPU time, time to first token, errors and cancelled, and peak parallel, above a daily tokens bar chart
The Usage page for the last 7 days. Example account with demo traffic. (tap to enlarge)
TileWhat it tells you
Input tokensEverything your tools sent to the model: your prompts, the files they read and the conversation so far. The line under it splits out how much was uncached.
Cache hit rateThe share of that input the model had already seen, so it was cheap to process again. Agents resend the conversation every step, so this is usually high.
Output tokensWhat the model wrote back, including any reasoning.
RequestsHow many generations ran, and on how many days.
GPU timeHow long the model spent working on your requests, not counting time spent waiting in the queue.
Time to first tokenHow long from sending a request to the first word coming back. The line under it shows average queue wait.
Errors · cancelledRequests that failed, and ones your tool cancelled.
Peak parallelThe most requests running at once, out of your plan's limit (for example 2 / 3), plus the most that were queued. See parallel streams.

Use 7D, 30D and 90D to change the range, or Custom for exact dates. The chart can show tokens, requests or GPU time per day. Export CSV downloads what you're looking at.

The Requests page

The Requests page with filters for key, model, endpoint, status and dates above a table of requests showing time, key name, endpoint, status, input, cached input, output, queue, time to first token, total time and client
Every request, newest first. Example account with demo traffic. (tap to enlarge)
  • Key: which API key made the call. If you gave each tool its own key, this tells you which tool it was. (See one key per project.)
  • Endpoint · model: the API route used (chat.completions, messages and so on) and the model.
  • Status: ok 200 for a normal response, or the error code.
  • Input / Cached input / Output: token counts for that request.
  • Queue: how long the request waited for a free stream. Near zero is healthy; a lot means you're at your limit.
  • TTFT and Total: time to first token, and the whole request's duration.
  • Client: the tool that sent it, as it identified itself.

Three things worth checking

  1. 1

    Why did that feel slow?

    Open Requests and look at Queue and TTFT for the slow request. A big queue means you were at your parallel limit; a big TTFT with no queue often means a very long prompt, which takes longer to read.

  2. 2

    Which tool is using the most?

    On Requests, filter by Key. With one key per tool, the filtered totals and the Client column show exactly who is responsible.

  3. 3

    Do I need more streams?

    On Usage, look at Peak parallel over a typical week. Sitting at the limit most days, or the banner telling you you're running into your parallel limit, means yes. Touching it rarely means no.

Are tokens billed?
No. The Usage page states it: tokens are never billed, and the page is for your visibility. Plans are priced by parallel streams.
Why don't I see my rejected requests on the Requests page?
Over-limit rejections (429) are counted on the Usage page rather than listed one by one on the Requests page.
Can I get this data into a spreadsheet?
Yes. Both pages have an Export CSV button. It exports the range and filters you currently have selected.
Sources