Google's SynthCorp generates verifiable data-agent benchmarks with up to 1,000 tables per estate

dair_ai · x · 2026-11-13

Google presents SynthCorp, a platform for building data-agent benchmarks that saturate less. It simulates routine and disruptive business events on a virtual calendar, materializing consistent relational tables, a document store, and a data catalog. Every operation lands in a simulation log, so gold answers are derived and verified without manual annotation. The first release ships seven data estates (up to 1,000 tables) and 256 tasks across five domains; multi-event composite tasks still trip state-of-the-art agents. Open-sourced for the community.

Original post →

More from Research

Research channel →