Rolling 24 hours
At every hour of the last week, the credits used in the 24 hours up to and including it. Hover or tap for the numbers.
Credits per hour, last 24 hours
The five most used models over 30 days, and all others together. Hover or tap for the numbers.
Requests per hour, last 24 hours
Models, last 24 hours
Every model, last 24 hours
| Model | Credits | Requests | Input tokens | of which cached | Output tokens | of which reasoning | Request share | First byte | Speed |
|---|
Models you can use
Every model AI credits buy. Put a name from a card in the model field of a request. Prices are credits per million tokens; 1,000,000 credits is 1 € of allowance. Use is over the window chosen above.
Loading the model list…
Top users, last 24 hours
The ten keys that used the most credits, each under an anonymous id. Other visitors cannot tell whose key an id is; the key's owner can work out their own (below).
| # | Id | Credits | Requests | Share of credits |
|---|
Find your own id
Your id is SHA-256 applied 1000 times: first to your key's label, a colon and the SHA-256 of your key, then each time to the previous result. A customer key's label is whmcs- followed by your client ID. In a terminal, with your key in ~/.virsilipasKey (the set-up line below puts it there):
REF=whmcs-12345 # your client ID here
k=$(sed -n 's/^key=//p' ~/.virsilipasKey | head -n 1 | tr -d '\n' | sha256sum | cut -d' ' -f1)
h=$(printf %s "$REF:$k" | sha256sum | cut -d' ' -f1)
for i in $(seq 2 1000); do h=$(printf %s "$h" | sha256sum | cut -d' ' -f1); done
echo "$h"
A new key (rotated) gives a new id. The list is rebuilt every hour.
Speed per model, last 24 hours
Time to the first byte of an answer, and output tokens per second while it streams. Speed shows once a model has a minute of measured streaming.
| Model | Requests | First byte | Speed |
|---|
When it is used
Credits by weekday and hour (UTC), last 24 hours.
Set up in one line
Any Linux terminal, as yourself, no root. It asks for your key (client area, AI key page), keeps it in ~/.virsilipasKey and points your AI tools at your credits. Run each as virsilipas-<name>.
wget -qO- https://virsilipas.mcx.fi/virsilipas.sh | bash
Add a tool's name to install it too:
wget -qO- https://virsilipas.mcx.fi/virsilipas.sh | bash -s -- copilot
Tools: codex, claude, gemini, opencode, pi, hermes, qwen, copilot, crush, goose, kilo, cline, kimi, vibe, aider, openhands, openclaw, zeroclaw, picoclaw, nullclaw, ironclaw, moltis, pikkumies.
Pikkumies, the Pulsed Media AI helper:
wget -qO- https://virsilipas.mcx.fi/virsilipas-pikkumies.sh | bash
No key yet? Pulsed Media services earn credits, or buy credits.
About this data
- Credits are what customers were charged. One euro of a customer's allowance is 1,000,000 credits.
- Input tokens include the cached part of the prompt; output tokens include the reasoning.
- Speed is output tokens per second while an answer streams: the output tokens of streamed answers that finished normally, over the time from their first byte to their end, across the time window chosen above. Answers that were not streamed are left out. It shows once a model has a minute of such time.
- Only Pulsed Media customer keys count; our own test keys do not. A request counts once it is accepted; one refused before that (no credit left, a rate limit, an unknown model) is not counted.
- Windows are complete UTC hours, up to the time shown above. The data is rebuilt every hour.
- Model names are the ones customers ask for, as listed with their prices on the AI credits page.
The data: usage.json, top.json and models.json (this site) ·pulsedmedia.com/data/v1/virsilipas.json · daily CSV · all Pulsed Media data feeds