Observed arrival · 2026-11-08
Debussy makes the quick calls AI agents need
An open decision model for AI agents that returns probabilities for safety judgments, tool routing, and whether it knows an answer.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Developers building AI-agent decision flows
- Worth noticing
- On HLE-MultiChoice, reported accuracy is 15.0%, below the listed 16.4% chance rate; the headline index is explicitly an estimate, not an official score.
Field notes
The page exposes three output families: safety judgments (harmfulness, refusal, harmful reply), tool routing (call, request details, or cannot answer), and knowledge choices that can include abstention. It lists variants with estimated memory needs of roughly 6–9 GB RAM and benchmark sets of 266–600 questions; the authors say settings were chosen on training data and test sets were used once.
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