a7580d324ad2

latestOpenAPI 3.1.0MITraw.githubusercontent.com2026-08-13977651.3 MB

Datasets

Cost per session by harness and model

Returns weekly refreshed, aggregated cost-per-session cells for the published harnesses. Sessions are never pooled across apps. Medians are of per-session USD spend, and privacy-preserving aggregation never exposes clerk_user_id values or per-session rows.

Filter by app_slug, model, or turn_range. Filtering by model alone works across apps for harness-vs-harness comparison at a fixed model. Results refresh weekly and include the source snapshot window in meta.

get/datasets/session-cost

Query parameters

app_slugstring

Filter to one published harness slug.

Example:hermes-agent

Filter to one published harness slug.

modelstring

Exact model permaslug filter. Works across all harness apps.

Example:anthropic/claude-4.8-opus

Exact model permaslug filter. Works across all harness apps.

turn_range'1-turn' | '2-9-turns' | '10-49-turns' | '50-plus-turns'

Filter by the inclusive number of turns in a session.

Example:10-49-turns

Filter by the inclusive number of turns in a session.

limitinteger

Maximum number of cells to return (1-500). Defaults to 100.

Example:100

Maximum number of cells to return (1-500). Defaults to 100.

offsetinteger nullable

Number of sorted cells to skip (0-5000). Defaults to 0.

Response

Aggregated cost-per-session cells for the requested filters.

Example response

{
  "data": [
    {
      "app_name": "Hermes Agent",
      "app_slug": "hermes-agent",
      "median_session_cost_usd": 1.74,
      "model_permaslug": "anthropic/claude-4.8-opus",
      "turn_range": "10-49-turns"
    }
  ],
  "meta": {
    "as_of": "2026-05-12T02:00:00.000Z",
    "window_days": 30,
    "window_end_date": "2026-05-11"
  }
}