About the project
Independent benchmarks for enterprise tasks
ERPBenchmarks publishes task-level observations to help readers understand what language models did under specific conditions—and where the evidence stops.
What is published
The current Adaptive Planning results are presented as one benchmark, Batch 1. They cover a limited question set, not every Workday process or tenant. The quality review is provisional and AI-assisted, not independently expert- or tenant-validated.
- Adaptive Planning · Batch 121 questions · 21 unique models
How to read a model score
A quality score on the shared 0–100 scale uses the first completed answer for each question and appears only after every question in the set has been assessed. Models without full assessed coverage remain unscored, not zero. There is no partial-set score or best-answer selection, and scores are not averaged across source runs.
Completion, latency, failures, and cost remain distinct. All actual attempts—including failures and attempts not selected for a question score—remain in the attempt and cost accounting; unknown cost is not zero. Source runs have differing request limits and question selections, so the graph is not an equal-compute trial or an aggregate efficiency ranking.
The expandable model rows use native HTML disclosure controls. They can be opened and closed by keyboard and work without JavaScript.
Independent, not official guidance
ERPBenchmarks is not affiliated with, sponsored by or endorsed by Workday, Inc. Workday and Workday Adaptive Planning are trademarks of Workday, Inc. Results should not be generalized to every tenant or treated as professional, implementation or purchasing advice.
A read-only site with no accounts
The site has no accounts, login, public submission forms or public write path. It does not run benchmarks from your browser and has no application analytics or tracking. Result details use native HTML controls and remain available without client-side JavaScript.
Published result content remains readable without client-side JavaScript. Hosting and network providers may process connection metadata under their own policies; this site cannot make promises about their handling of it.