Yuetai Li, a PhD student at the University of Washington, who is working with Prof. Radha Poovendran and ACTION undergraduates Brian Lee from the University of Chicago and Kayla Xu from Northwestern, helped build two AI evaluation benchmarks: JobBench and Agents' Last Exam. Today, both of those benchmarks were name-dropped in the same-day flagship model launches from Meta and OpenAI.
When Meta released their agentic/coding-focused model, Muse Spark 1.1, they noted how their new model tops the JobBench benchmark. In other words, Meta used JobBench as one of the key yardsticks to show off how good their new model is at doing realistic professional tasks.
OpenAI also launched GPT-5.6 (with three tiers: Sol, Terra, Luna) today. In their release blog, they highlight performance on Agents' Last Exam, describing it as an evaluation of long-running professional workflows across 55 fields, where GPT‑5.6 Sol sets a new high of 53.6.
Getting one of our benchmarks adopted by a frontier lab is a strong form of validation. Having it happen with two different benchmarks, on the same day, from two rival labs is a milestone worth celebrating.