A 395M label-only bidirectional encoder achieves 82.90 average F1 on moderation benchmarks without reasoning, matching larger reasoning decoders at ~100x lower inference cost and with better robustness properties.
arXiv preprint arXiv:2606.16902 , year=
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Do Safety Guardrails Need to Reason? LeanGuard: A Fast and Light Approach for Robust Moderation
A 395M label-only bidirectional encoder achieves 82.90 average F1 on moderation benchmarks without reasoning, matching larger reasoning decoders at ~100x lower inference cost and with better robustness properties.