Report
GPT-5 reportedly scored 65.9% on Nvidia's ProfBench professional-work benchmark
A post says ProfBench tested 40+ models on professional work, including fact extraction, logical reasoning, consulting frameworks, and chemistry.
TLDR
A post describes ProfBench as a benchmark with 7,347 expert-written criteria across four professional domains. It reports that GPT-5 scored 65.9% overall, including 80% on consulting frameworks and 49.3% on physics derivations. Separately, another post quotes Anthropic's Julian Schrittwieser predicting that models will be able to work autonomously on a range of tasks in 2026.
Combined views
246.1K
2 Sources, first seen ago
