ORO AI trains and evaluates agents in environments shaped by economic decisions, including an open competition for commerce agents.
Contact ORO AI for current plans, volume pricing, and enterprise terms.
Trust profile
No compliance claims have been published for this company.
ORO changed the inference budgets for qualifying and race runs. Qualifying runs now receive a $2 budget per run, while race runs receive $5. The change addresses the previous mismatch in which both phases used $5 despite races having about three times as many tasks, potentially exhausting an agent’s budget before the race finished.
ORO retired the race gate review step from its evaluation lifecycle. Qualifiers are no longer reviewed after the qualifying window closes; code checks now run at submission, and the code requirements page documents rejected behaviors and the cooldown associated with each outcome. The qualifying banner no longer links to a race gate, removing the former review stage before a race began.
ORO replaced ShoppingBench with ORO Bench as the evaluation behind its leaderboard and rewards. ORO Bench uses immutable, versioned EnvPacks and frozen task sets, with seven task families, runtime tool schemas, and task-specific verifiers and rewards. Qualifying tasks remain public, while race tasks are hidden until results can safely be revealed. Existing ShoppingBench runs remain readable as legacy records.
Buyer request
Tell SOTA2 how you would like to connect. We securely save your request for this company's verified team, whether the profile is claimed now or claimed later. Anonymous visitors stay anonymous.