List your company

Tool-use safety validation on safety-stress set 80-task

0Harm Rate (Call)RACG
-15.7512.519.25Jun 11, 2026

Evaluation results

MethodLinks
RACGModel=Claude Opus 4, lambda=22026.06001002,265
RACGModel=Claude Sonnet 4.6, lambda=22026.06001002,722
RACGModel=Claude Haiku 4.5, lambda=22026.06001002,718
RACGModel=GPT-OSS 120B, lambda=22026.06001001,242
RACGModel=Nova Premier, lambda=22026.06001002,910
RACGModel=Nova Pro, lambda=22026.06001001,800
RACGModel=Nova 2 Lite, lambda=22026.06001003,534
CMTFModel=Claude Opus 4, Variant=causal2026.0625251001,936
CMTFModel=Claude Sonnet 4.6, Variant=causal2026.0625251002,329
CMTFModel=Claude Haiku 4.5, Variant=causal2026.0625251002,326
CMTFModel=GPT-OSS 120B, Variant=causal2026.0625251001,065
CMTFModel=Nova Premier, Variant=causal2026.0625251002,490
CMTFModel=Nova Pro, Variant=causal2026.0625251001,539
CMTFModel=Nova 2 Lite, Variant=causal2026.0625251003,025