SMB Bench v1.0 research report

Website

• The new SMB Bench v1.0 research report asks whether AI can learn from the delayed consequences of its own decisions while running a wholesale buying desk for a 90-day quarter. Nine frontier models are scored on the profit they make, against a fixed buying rule.
• SMB Bench replaces the ICCA Bench report, and the old report's address now opens SMB Bench.
• The other research reports read more easily, with shorter figure captions, clearer charts and tables, and layouts that fit phones.