FrontierThe story, in brief
New benchmark exposes how badly AI struggles with real knowledge work
3%. That's the full solve rate on realistic knowledge work — even for the best AI models today.
The KeyNews take
Why it matters
A new benchmark reveals a critical capability gap between lab performance and real-world knowledge work, forcing a reckoning on how founders and enterprises should assess AI readiness for mission-critical workflows.
The key facts
4 to knowBest-in-class AI models solve only 3% of realistic knowledge work tasks fully
Benchmark measures real-world knowledge work performance (not synthetic tasks)
Published June 2026
Suggests major gap between benchmark claims and production deployment viability
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: Even the best AI model fails at realistic knowledge work, fully solving just 3 percent of tasks.