Harder Data, Same Agent: How GPT-6 Astra Got to the Top
GPT-6 Astra is the only one of eight models that rose from OfficeQA v1 to v2, 72.2% to 73.3%, and took first place on the hard set at $11.46 per question against Fable 5.1's $1.24. It went through the same Energent agent on both Databricks OfficeQA editions the day it reached the API.
Read article

