Observed arrival · 2026-11-19
Abstend checks whether local models know when to stop
Abstend publishes scorecards testing open AI models’ factual knowledge and whether they admit uncertainty across specific files, quantizations, and server settings.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Local-LLM users and model evaluators
- Worth noticing
- Each configuration is tied to a file hash, build flags, sampling details, a 95% interval, and a public receipt.
Field notes
The test separates easy facts, hard but real facts, and questions about invented entities, then tracks correct answers, declines, and refusals separately. The page says each scorecard identifies a specific file and configuration, while comparisons are preregistered and base models are replayed alongside variants. A dated corrections log records changes to measurements and their stated reasons.
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