Usage Explorer

Chart cost, tokens, requests and latency grouped by model, key or time, with filters and shareable URLs.

Usage Explorer at /dashboard/usage charts your AnyRouter activity. Every view is encoded in the URL, so a shared link opens the same chart.

A chart combines a dimension (what rows are grouped by, one series per group), a metric (the y-axis), a granularity (bucket width) and filters (applied before grouping).

Dimensions

DimensionOne series perUse it for
timeBucket (hour, day, week)Trends
modelModel idAttributing spend to models
keyAPI keyIsolating cost or errors to a key
organizationOrg or workspacePer-team cost
routeAPI path (/chat/completions, /responses, /embeddings)Traffic split by endpoint

Dimensions stack with filters: group by model while filtered to one key to see that key's model mix.

Metrics

MetricUnitNotes
CostUSDPass-through provider cost, no AnyRouter markup
Prompt tokenstokensIncludes cached prompt tokens
Completion tokenstokensGenerated output
RequestscountOne per request
p50 / p95 / p99 latencymsServer-measured, end to end, computed per bucket so incidents show quickly
Error ratepercentNon-2xx divided by total requests in the bucket
Cache hit ratiopercentcached_tokens / prompt_tokens; 0.85 means 85% of input tokens were cache reads

Granularity

GranularityBest rangeUse for
HourLast 24-72 hoursSpikes and incidents
DayLast 7-60 daysWeekly cost reviews
WeekLast 60+ daysQuarterly trends

Filters

Filters are AND-composed: date range (start and end, or Last 24h, Last 7d, Last 30d, Month to date), model, API key (by sk-ar-... prefix), organization or workspace, route (/chat/completions, /responses, /embeddings, /messages), status (2xx, errors or both), and cached or uncached.

https://anyrouter.dev/dashboard/usage?granularity=day&dimension=cost&group_by=model&from=2026-04-01&to=2026-04-30&models=openai/gpt-5.4,anthropic/claude-sonnet-4-6

To compare two keys or models on identical traffic, set group_by=key (or model) and add both ids to that filter.

Bars suit discrete buckets and lines suit trends; the chart picker switches without rebuilding the query. Grouped series stack by default; switch to side-by-side to compare series. Zero-value buckets stay on the axis, so a missing bar means no requests, not missing data.

Export and drill down

Chart CSV export is coming soon. Today, export the underlying requests with GET /api/v1/logs?format=csv, which accepts the same date range, model and key filters (see Request Logs). Click a bar to open that slice in Request Logs.

Troubleshooting

SymptomFix
Chart is emptyWiden the date range (default Last 30 days), clear filters one at a time (they AND together), and check the workspace switcher: a workspace you cannot access returns zero rows, not an error.
Total differs from invoiceThe chart shows request-level cost at request time. Invoices include adjustments (credits applied, free-tier usage, refunds); see Credits.
Cost formulaprompt_tokens × input_price + completion_tokens × output_price using the model page prices; cache reads use the upstream's cache-read rate.
What counts as cachedThe upstream reported usage.prompt_tokens_details.cached_tokens (Anthropic-style breakpoint or OpenAI prefix caching). See Prompt Caching.

AnyRouter routes requests to third-party models. Models and outputs are provided as is, with no warranty; you use them at your own risk. Disclaimer