Benefit Street Partners Deploys GPT-6 Astra for 300-Page Credit Analysis as Review Shrinks to 20 Minutes
Updated
Updated · OpenAI · Oct 16
Benefit Street Partners Deploys GPT-6 Astra for 300-Page Credit Analysis as Review Shrinks to 20 Minutes
2 articles · Updated · OpenAI · Oct 16
Summary
Benefit Street Partners has put a GPT-6 Astra-based credit agreement analyzer live on its internal portal, giving teams a first-pass read of loan documents in about 20 minutes.
The tool targets agreements often running 300 pages or more, flagging clauses that could shift value from lenders before analysts spend an hour or more on human review.
Codex is also speeding underwriting rounds: portfolio manager Anant Kumar said follow-up questions that once took days or weeks can now be turned in a day, and often answered during committee meetings.
Kumar also used AI-assisted coding to build a BDC valuation tracker from SEC filings—first version in a day, roughly 10 to 12 hours total—letting BSP compare its loan marks with other holders'.
BSP says the aim is not to do more deals but to add review capacity and rigor within the same timeline, with longer-term plans to connect data across deals for continuous underwriting.