The Guardrails tab surfaces enforcement from your prompt injection and PII rules. More on how Guardrails work here:

On Overview you get the whole picture at once: usage by model, BYOK vs OpenRouter spend, request volume by model, token breakdown, and prompt caching, all stacked on a single screen.

The Trends tab breaks spend down by model over time, with a Trending panel showing what's rising and falling across the catalog.

Last week GLM 5.1 spiked >999% for our internal use, with a new entrant this week: Claude Fable 5!

Model neurodiversity, visualized:

Notice that prompt caching card.

Our cache hit rate sat at 82.8% last week, with caching peaking above 8B tokens on Jun 9. Click any card on Overview to go deeper...

...which drops you into Explore, a fully dynamic analytics view.

Here it's Cached Tokens by Model: Sonnet 4.6 alone accounts for 16.2B cached tokens, just under half the total.

Flip the same view to Users and you can see who's driving spend across your org.

One OpenRouter engineer climbed >999% week over week while another dropped 99%. You can catch runaway usage long before it reaches your invoice.

You can also group by agent. Claude Code leads here, with Cursor showing up as a brand new entrant.

Every coding agent and client your org touches, ranked by spend over time.

Explore is yours to shape. Pick any Metric: total usage, request count, tokens by prompt, completion, reasoning, or cached, and latency all the way down to P50, P90, and P99.

Then group by whatever matters: Model, Provider, API Key, Origin, Country, and more. Slice your traffic from any angle.

Roll it up over any window: Hourly, Daily, Weekly, Monthly, or Total.

Build the precise view you want and watch it update as traffic flows in.

Available now for everyone! Open yours: https://openrouter.ai/activity

@OpenRouter cache hit rate is very nice and useful!

maybe can get rid of the 0s for missing data, like claude fable 5 being 0 on days before its launch is clearly a case of missing data, not actually 0% cache hit.