Not Specifying Model Versions Is Dangerous: A Case Study of 40 Models in Production

https://noon-wiki.win/index.php/Cut_to_the_Chase:_What_the_GPT-5.2_xhigh_Mode_10.8%25_Hallucination_Claim_Really_Means_for_Enterprise_Document_Workloads

This case study documents a real-world evaluation run by an ML operations team preparing a conversational assistant for enterprise use