Usage & Insights

See how Aivana MMI routes intelligence, performs at scale, and compares economically with leading AI benchmarks.

Usage & Insights shows how your applications use Aivana MMI over a period you choose: 7, 30 or 90 days, or a custom range of up to 90 days. One date control drives every figure and chart on the page, and each figure is compared with the period of the same length just before it.

• Total MMI runs: completed runs you were charged for.
• Customer input and output tokens: the tokens you were charged for, with cached input included in input.
• MMI billing: what was deducted from your usage credit balance. Promotional tokens are not counted here; their usage is on the Billing page.
• Estimated savings vs Frontier Blended: an estimate of how your cost compares, shown once your usage is large enough. See Benchmarks.
• Average run latency, with P50 (median) and P95 by day.
• Intelligence routing mix: how many runs used Efficient, Advanced or Premium intelligence.
• Knowledge source: how many runs answered from Native Intelligence and how many used Live Web Search.
• Every chart can also be shown as a table.

What you paid is actual. The Frontier Blended benchmark is an estimate: what the same customer-visible workload would cost at representative pricing for leading frontier-model APIs, compared across the whole period. Usage covered by promotional tokens counts as free, so it raises the savings. Turn on Exclude promotional tokens to leave it out and compare only the usage you paid for with credits. When your total is lower it reads Estimated savings; when it is higher it reads Estimated cost difference and says how much higher.

The comparison unlocks once the period has at least 100 comparable runs or 1,000,000 comparable tokens with at least 90% coverage, and until then the page shows how far along you are. When only part of the period could be compared, it says so. Benchmarks explains how the estimate is calculated, what it leaves out, and why some usage is not compared.

Aivana decides how much intelligence each request needs. Efficient suits straightforward requests, while Advanced and Premium apply progressively more depth to harder ones. The routing mix shows how your requests were spread across the three levels. It is a summary of outcomes and does not describe how an answer was produced.

A run counts as Live Web Search only if a search actually ran and returned sources, not merely because search was available. Runs recorded before source tracking began are not classified, so they are left out of this split rather than assumed to be Native Intelligence. When a period includes them, the chart says so.

Days are cut at midnight UTC everywhere on this page, including the charts and the date range. A run made late in the evening in your own time zone may therefore appear on the next day.

Choose View usage details for the run-by-run list and per-key activity. Each run is identified by the request_id your API responses return, so you can find the run behind any response. Cached input is shown beside input tokens and is part of it. A run you stopped shows as stopped, and is charged for the prompt and what you received.

GET /v1/metrics returns token totals for a recent period, and GET /v1/quotas your current usage against your limits. Both need a key with the metrics scope, which a key created without scopes has.

The period for GET /v1/metrics is counted in whole UTC days, including the current UTC day: days=7 means today and the six days before it. Every response says whether it is complete. truncated is false when the figures cover the whole period. If your workspace has more activity in the period than one request can read, truncated is true and the oldest days that could not be read are left out, never estimated. complete_from then gives the oldest UTC day (YYYY-MM-DD) from which every day through today is counted in full; it is the first day of the period when nothing is partial, and null only if even today is partial. Treat complete_from as the start of the figures you can rely on.