Observed arrival · 2026-11-21
AktenBench puts language models through full German case files
A benchmark compares how language models extract deadlines, amounts, and supporting evidence from complete German legal files.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Legal AI evaluators and model developers
- Worth noticing
- The benchmark reports 62.8% of pages as irrelevant or misleading material, and tests each file as scan PDF, digital PDF, and text.
Field notes
The benchmark describes 14 fictional but realistic case files across 14 legal fields, with each file available as a scan PDF, digital PDF, and text. Its scoring checks deadlines, amounts, and evidence against a final answer sheet, and the leaderboard breaks results out by file length and task area. The project says only eight of twelve listed models have initial results; it also flags the scores as provisional.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
One card from the complete issue