FDE-Bench covers 29 combinatorial optimization and 20 machine learning cases.
AI agents still cannot reliably deliver end-to-end on underspecified business requests.
New benchmark of real customer projects finds best agent configuration scores 0.55 out of 1.0 against accepted human solutions.
Independent
Huaiming Li · Can Huang · Chufan Wu · Danyu Liu · Xiaomin Yuan · Qinze Li · +15 more
Research Digest··2 min read
Thread:Process Agent Benchmarks
The authors present FDE-Bench, a benchmark of 49 cases built from real customer projects.
Why this paper
Independent · Part of Process Agent Benchmarks, now 27 papers
In one line
FDE-Bench tests agents on real business requests needing clarification and optimization; best agent achieved 0.5510 delivery score versus a 1.0 FDE reference.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§