Task overview
Summary JSON ↗| Task | Success rate | Input tokens | Output tokens | Total tokens | Usage coverage |
|---|
One seed per subtask · Native success criteria. Input includes cached tokens; output includes reasoning tokens. Total = input + output.
≥ indicates known lower bounds where usage is incomplete. Rates exclude interrupted and incomplete executions. This snapshot combines retained historical outcomes with completed instruction revisions. Both LIBERO results retain their measured Kinex 0.8.0 identity.