List your company

Reward Hacking Onset Localization on CHERRL VerInstruct and HealthBench runs

490VerInstruct SPCC-Qwen
325.68368.34411453.66Jun 3, 2026

Evaluation results

MethodLinks
CC-Qwenframework=Claude Code, backbone=Qwen2026.064902209691341474198800
CC-Haikuframework=Claude Code, backbone=Haiku2026.064901501001013311584203290
RHDA-397Bbackbone=qwen3.5-397B-A17B2026.064891577683385459167200
RHDA-Plusbackbone=Qwen3.5-plus2026.064821328675383454120110
Reference2026.064781169168301460———
CC-Opusframework=Claude Code, backbone=Opus2026.06470151110901214502742240
CC-Sonnetframework=Claude Code, backbone=Sonnet2026.064632189368437446269860
CoT Monitor2026.06332169——283—2171723