Wire · opportunities
'Far from human level': AI models score below 25% on real-world job tasks, UC Berkeley study finds
◆ Sectors
◆ Geography
◆ Source
◆ Verified
Fusion42 · 20 July 2026 · Fusion42 review
UC Berkeley's real-world professional task benchmark found leading AI models (including GPT-5.5) score below 25% on complex workflows across 55 industries, with 0% success on the hardest tier. The study shows current AI excels at repetitive tasks but struggles with sustained reasoning, execution reliability, and adaptive problem-solving.
This Wire brief sits within Fusion42's coverage of AI Frontier Models.
◆ ◆ The Wire takeaway
If you're building AI tools for professional workflows, you're still selling belief, not capability: the models fail at the complex, adaptive work that actually drives revenue in finance, law, and manufacturing. Build for the 25% of tasks that are routine and repeatable—that's where AI delivers today, and where job displacement will actually happen first.
◆ Coverage
1 source · 20 Jul 2026
◆ Related on Wire
◆ Topics