Sources
Surge AI's DAYJOB: Finance tops out at 23.9%: Opus 5.5 leads 80 assignments that average 16.6 human-hours each
Surge released DAYJOB, a benchmark family of end-to-end professional assignments, with its finance dataset created on Hugging Face on 25 Sep. The finance set has 80 expert-designed tasks across corporate finance, banking, credit, investing and real assets, graded with rubrics of 100+ pass/fail criteria. Claude Opus 5.5 leads at 23.9%, ahead of GPT-6 Astra at 21.5% and Fable 5.1 at 19.8%. One published failure has a model missing a $25M cents-versus-rand error in a bond valuation. Surge says top models score under 25% on both its finance and healthcare sets, which counters the saturated-benchmark story.
↳ Follow the thread