Moneda Bench, made by Moneda

Moneda FilingsBench

Questions about a US company's SEC filings with no document attached. The system has to find the right filing and period itself, then read the figure: reported lines, ratios with the formula given, fiscal years that do not end in December, figures as originally reported before a restatement, and comparisons.

Best reported

91.6

Numa 1Moneda

% correct (private split)

Run by Moneda, 2 Nov 2026

About this benchmark

Size
About 400 questions a quarter from 100 US companies, built only from filings made after a cutoff date
Public split
Yes, about 1 question in 10; the rest are private and only those are ranked
Grading
Code against the figures in the filings' XBRL data, as first filed, with stated tolerances (0.5% for amounts, 1 cent per share, 0.1 point for percentages). A small set of questions on reportable segments is graded by two judges against a rubric.
Licence
Items CC BY 4.0, harness MIT
Submission
Open harness at release
Products allowed
Yes
What it does not cover
Filings outside the US, documents handed to the system, and analysis or judgement beyond reading the filing. Systems with no search or tools are not scored.

How it runs

Answer keys are computed by code from SEC companyfacts, using the value from the first filing that reported it. Each quarter the set is rebuilt from filings made after a new cutoff, so the answers are newer than the models tested. About 1 question in 10 is public; the rest are private and only those are ranked.

Status: first results published 2 Nov 2026.

Results

1 models
Leaderboard

Each score shows who reported it: the benchmark's authors (quoted from their paper), the benchmark's official board, or Moneda (run with the benchmark's own grader and rules). We do not copy other aggregators.

Moneda FilingsBench

% correct (private split), higher is better. 1 models.

Updated 2 Nov 2026

  1. Numa 191.6

Moneda FilingsBench

% correct (private split), higher is better. 1 models.

Updated 2 Nov 2026

  1. 1Numa 191.6
Moneda FilingsBench leaderboard (% correct (private split), higher is better)
1Numa 1Moneda91.6361 private questions, no document given; graded by code against the filings' XBRL data, 13 segment questions by a two-judge panel (not yet human-audited)Moneda run, 2026-11-02
Run by Moneda, 2 Nov 2026

Paper, data and harness will be linked here at release.

Last checked 1 Nov 2026. Something out of date or wrong? Send a correction.