Sources / Press and newsletters
Sequoia Capital podcast, Brendan Foody (Mercor): How RL environments are built
4 records were read from this source, dated 15 Oct 2026.
Its permanent copy is being made in the Wayback Machine.
Last checked 15 Oct 2026
What was read from it
- KindPrice observationAs statedabout $500K in compute per post-training run on 1,800 tasksDated15 Oct 2026
- KindPrice observationAs stated$2,000 per taskDated15 Oct 2026
- KindPrice observationAs stateda $1,500 cost basis per taskDated15 Oct 2026
- KindPrice observationAs statedfrom $50 to $10,000 per taskDated15 Oct 2026
Figures it feeds
- $500K per datasetDatasets, tasks and environments, cost to produce, per dataset
- $1,500 per taskDatasets, tasks and environments, cost to produce, per task
- $2,837 per taskDatasets, tasks and environments, made to order, per task
- 9.5xHow much more buyers pay than makers are paid for datasets, tasks and environments, per task
- 36Prices per task behind every valuation of a task set