ApprenticeBench: Agents Continually Learn Real Jobs, Surpassing Human Pros

ysu_nlp · x · 2026-09-12

NeoCognition introduced ApprenticeBench, a benchmark combining computer use with continual learning on real jobs. The team reports Fable 5.1 and GPT-6 Astra can continually learn on the job and surpass human professionals — a "decisive step change" in AI job readiness. Notably, no forward-deployed engineers are involved: agents deploy themselves into the role, testing the full zero-to-productive loop.

Related event: NeoCognition Launches ApprenticeBench, a Job-Level Continual Learning Benchmark(12 posts)→

Original post →

More from coding & agent

coding & agent channel →