A $500 RL fine-tune of a 9B open model beat frontier models on catalog review

# Summary Researchers fine-tuned a 9-billion parameter open-source model using GRPO for $500 and achieved better performance on catalog review tasks than frontier AI models, costing 40-340× less per 1,000 listings. The article argues that companies seeing the greatest returns on AI investment are those redesigning entire workflows around AI capabilities rather than simply integrating models into existing processes, with workflow redesign being the most correlated attribute to financial impact according to McKinsey research.