Not Missing a Single Detail: Hebbia's Financial AI Gets a 20% Accuracy Boost With Claude Fable 5

Hi everyone, it's Shiichan! Today's story is about how far AI can be trusted in finance, a field where missing even one detail just isn't an option.

Claude Blog claude.com

What was announced?

Another "Working at the frontier" post on the Claude Blog covers Hebbia, which builds an AI platform for financial due diligence, and how they're putting Claude Fable 5 to work.

According to Divya Mehta, Hebbia's founding product manager, Fable 5's biggest strength is "the ability to find the right information from a dense data set, and then synthesize it correctly." Researcher Joe Renner says it's the best performance he's measured across any new model.

The story so far

With earlier models, complex multi-part requests would sometimes end up with a few questions quietly dropped along the way. In finance, where decisions ride on getting every detail right, that kind of gap can be a serious problem.

What changes

Claude Fable 5 can hold onto complex, multi-part requests at once without dropping any of them. Rather than inferring an answer, it grounds each claim in the actual source documentation.

On question-answering and citation tasks over financial documents, that translated into roughly a 20% relative accuracy improvement, the largest accuracy gain Hebbia has recorded in finance-specific testing.

Hebbia's platform is used by more than a third of the top 50 asset managers, plus tier-1 investment banks and law firms. Their benchmarks compare against whatever model they're about to replace, and the bar is set intentionally high since customers rely on it for deal-level decisions.

Dive Deep

Tasks like building an investment deck, which used to take 2-3 days, can now be compressed down to minutes. That frees up bankers to focus on strategic positioning instead of compiling data.

Wrap-up

• Hebbia, a financial due diligence AI company, recorded its largest-ever accuracy improvement with Claude Fable 5
• About a 20% relative accuracy gain on question-answering and citation tasks over financial documents
• Complex multi-part requests no longer get dropped, and claims are grounded in source documents
• Tasks like investment deck creation went from 2-3 days down to minutes in some cases

Worth watching if you're curious how far AI can go in fields like finance and law, where there's zero room for missed details.