Office work and professional documents
Producing the finished files a professional hands over: reports, slide decks, spreadsheet models and memos. Each test here judges those files its own way: against professionals' own work, against criteria written by experts, by blind comparison with other AI work, or on whether the result is correct.
Sources last checked 2026-10-23.
We can't show scores for this kind of work yet. These testers publish results: Scale AI and Center for AI Safety (one test), Mercor (one test), Artificial Analysis (one test), Vals AI (one test) and Renmin University of China (RUC KB Reasoning lab) (one test). See who tests it
At a glance
Vendor claims have not been collected for any category yet.
ⓘ More about this category
- Reliability not yet measured
- 5 sources, linked results only
- Sources last checked 2026-10-28
- Ring 4: needs rings 1 to 3 first
- RLI = Remote Labor Index (RLI), APX = APEX-Agents, GDP = GDPval-AA v2.1, EMB = Excel Modeling Benchmark v1.1, SB2 = SpreadsheetBench 2
Tested by
Remote Labor Index (RLI)
Scale AI and Center for AI Safety, checked 2026-10-28The share of real paid freelance projects where an AI agent's deliverable is judged by expert reviewers to be at least as good as the professional's own work.
Scale AI reports that failed projects cluster around fundamental issues, such as poor quality.
Last updated 2026-10-22Covers 14 of the models we trackScale AI says the models at the frontier run on its data.Scores not shown: permission requested on 2026-10-27
See their resultsMore about this test: Remote Labor Index (RLI)
Reliability not measured by this source.
Technical details for Remote Labor Index (RLI)
Automation Rate: The share of real paid freelance projects where an AI agent's deliverable is judged by expert reviewers to be at least as good as the professional's own work.
License: Score reuse not stated. Checked 2026-10-28.
APEX-Agents
Mercor, checked 2026-10-28The average share of expert-written grading criteria an AI agent meets on long, multi-application tasks from investment banking, management consulting and corporate law.
Last updated: not published by the tester. We last saw current models listed on 2026-10-28.Covers 27 of the models we trackMercor licenses datasets built by its expert network.Scores not shown: permission requested on 2026-10-27
See their resultsMore about this test: APEX-Agents
Reliability not measured by this source.
Technical details for APEX-Agents
Mean Score: The average share of expert-written grading criteria an AI agent meets on long, multi-application tasks from investment banking, management consulting and corporate law.
License: Score reuse not stated. Checked 2026-10-28.
GDPval-AA v2.1
Artificial Analysis, checked 2026-10-28A relative rating, from blind side-by-side comparisons, of how good an AI agent's work products are on professional tasks such as documents, slides and spreadsheets across 44 occupations.
Last updated: not published by the tester. We last saw current models listed on 2026-10-28.Covers 34 of the models we trackOpenAI developed its tasks, in collaboration with industry professionals.Scores not shown: their terms forbid republishing
See their resultsMore about this test: GDPval-AA v2.1
Reliability not measured by this source.
Technical details for GDPval-AA v2.1
Elo: A relative rating, from blind side-by-side comparisons, of how good an AI agent's work products are on professional tasks such as documents, slides and spreadsheets across 44 occupations.
License: Terms of Use forbid republishing. Checked 2026-10-28.
Excel Modeling Benchmark v1.1
Vals AI, checked 2026-10-28How well an AI agent builds working Excel financial models, such as buyout, valuation and merger models, judged on numbers, formulas and presentation against expert-built models.
Vals AI reports that buyout and discounted cash flow models are long chains of linked calculations, so errors spread through the whole model.
Last updated 2026-10-26Covers 9 of the models we trackScores not shown: permission requested on 2026-10-24
See their resultsMore about this test: Excel Modeling Benchmark v1.1
Reliability not measured by this source.
Technical details for Excel Modeling Benchmark v1.1
Accuracy: How well an AI agent builds working Excel financial models, such as buyout, valuation and merger models, judged on numbers, formulas and presentation against expert-built models.
License: Score reuse not stated. Checked 2026-10-28.
SpreadsheetBench 2
Renmin University of China (RUC KB Reasoning lab), checked 2026-10-28How often an AI agent correctly completes whole business spreadsheet jobs, such as finishing financial models, fixing errors in workbooks and building charts.
Last updated 2026-09-27Covers 3 of the models we trackIts contributors include Shortcut.AI, which sells a finance agent.Scores not shown: no license for reuse stated
See their resultsMore about this test: SpreadsheetBench 2
Reliability not measured by this source.
Technical details for SpreadsheetBench 2
Overall accuracy: How often an AI agent correctly completes whole business spreadsheet jobs, such as finishing financial models, fixing errors in workbooks and building charts.
License: Score reuse not stated. Checked 2026-10-28.
What this doesn't tell you
- None of these tests used your templates, your clients or your quality bar.
- Each tester judges quality its own way, so scores from different tests do not compare.
- Scores from different sources are not comparable, even when they share a unit.
- None of these tests were run on your own documents, code or data.
- No source for this task allows us to show its figures yet. Follow the links for their results.
Data version 2026-10-28+91ac3fc9d071 · Terms of use