Simulations

How AI agents and models did in published tests: an agent that runs a vending business for a year, one that handles refund requests from simulated customers, one asked to attack a test network. Plain words first, then the set-up and the results for specialists, with a chart and a table that compare the systems.

TOSWO Research does not run these tests. The figures are the testers' own, taken from their published reports, and every figure in a table is matched to a passage of the report. A test shows how systems behaved in one scenario, not how good they are in general.