Datasets:

Formats:
json
Size:
< 1K
ArXiv:
License:
HaoyuDong commited on
Commit
dcef7a1
·
verified ·
1 Parent(s): 6300ad1

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +1 -1
README.md CHANGED
@@ -57,7 +57,7 @@ We adopt a three-step workflow labeling process:
57
 
58
  This process yields **172 enterprise-grade workflows—primarily multi-task composite workflows**, involving 1,710 spreadsheets and 27 million cells, capturing the intrinsic **messy, long-horizon, knowledge-intensive, and collaborative nature** of real-world finance & accounting work. In this release, we provide full annotations for the first 72 workflows, with the remaining 100 to be released in a subsequent update.
59
 
60
- <img src="figs/distribution_chart.jpeg" width="1000" />
61
 
62
  We conduct both human and automated evaluations of frontier AI systems including GPT5.1, Claude Sonnet 4.5, Gemini 3 Pro, Grok 4, and Qwen 3 Max, and GPT 5.1 Pro spends 48 hours in total yet passes only 38.4% of workflows, while Claude Sonnet 4.5 passes just 25.0%, revealing a substantial performance gap for real-world enterprise scenarios.
63
 
 
57
 
58
  This process yields **172 enterprise-grade workflows—primarily multi-task composite workflows**, involving 1,710 spreadsheets and 27 million cells, capturing the intrinsic **messy, long-horizon, knowledge-intensive, and collaborative nature** of real-world finance & accounting work. In this release, we provide full annotations for the first 72 workflows, with the remaining 100 to be released in a subsequent update.
59
 
60
+ <img src="figs/distribution_chart.png" width="1000" />
61
 
62
  We conduct both human and automated evaluations of frontier AI systems including GPT5.1, Claude Sonnet 4.5, Gemini 3 Pro, Grok 4, and Qwen 3 Max, and GPT 5.1 Pro spends 48 hours in total yet passes only 38.4% of workflows, while Claude Sonnet 4.5 passes just 25.0%, revealing a substantial performance gap for real-world enterprise scenarios.
63