Skip to main content

Usage

The spend of the calling project.

📄️Read the speed and the reliability of each model

Returns, for a window and for each model, the answered share, the p50, p90, p95 and p99 time of the answered calls, the p50 and p95 time to the first token, the speed in tokens per second, the cost in `cost_usd` and in `cost_millicents`, and `total`, which is how many models the answer holds. `answered` is the number of answered calls and `cost_usd` is the cost in United States dollars, one hundred thousandth of which is one millicent. A time to the first token and a speed are null when no answered call was streamed, and `speed_calls` says how many calls they cover.

📄️Read the spend of a window and of the window before it

Returns the totals of a window: the cost in `spent_usd`, `paid_usd`, `plan_usd` and `list_usd`, the calls in `requests`, the tokens, the attempts, the answered calls, the retries, and `cost_not_known`, which is how many calls carry no cost and no price. `paid_usd` is the part of the bill that the project paid, `plan_usd` is the part that its plans covered at the list price, and `list_usd` is their sum. `previous` holds the cost, the calls, the tokens and `cost_not_known` of the window of the same length that ends where this one starts, so a reader states the change between the two. `from` and `to` name the times that the figures cover. The answer also holds `by_model`, `by_tier`, `by_key`, `by_source`, `by_key_source` and `series`, which split the same window.