Claude Opus 5 (Low): benchmark results

Provider: Anthropic. Released 2026-07-24. Access: API.

Aggregate Rating 1846 ± 9, rank #81 of 2170 rated models, from 161 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
NTBench - AgentDojo Near Target Harmful ASR@11.4Attack success rate at 1 (%)100
NTBench - AgentDojo Near Target Harmful ASR@32.8Attack success rate at 3 (%)100
NTBench - AgentDojo Near Target Incorrect ASR@13.4Attack success rate at 1 (%)100
NTBench - AgentDojo Near Target Incorrect ASR@37.5Attack success rate at 3 (%)100
NTBench - AgentDyn Near Target Incorrect ASR@16.2Attack success rate at 1 (%)100
NTBench - AgentDyn Near Target Incorrect ASR@316.5Attack success rate at 3 (%)100
NTBench - Dolly: Closed QA Near Target Harmful ASR@10.2Attack success rate at 1 (%)100
NTBench - Dolly: Closed QA Near Target Harmful ASR@30.5Attack success rate at 3 (%)100
NTBench - Dolly: Extraction Near Target Harmful ASR@10.2Attack success rate at 1 (%)100
NTBench - Dolly: Extraction Near Target Harmful ASR@30.5Attack success rate at 3 (%)100
NTBench - Dolly: Summarization Near Target Harmful ASR@10.2Attack success rate at 1 (%)100
NTBench - Dolly: Summarization Near Target Harmful ASR@30.5Attack success rate at 3 (%)100

Interactive version: theaggregate.ai/model?slug=claude-opus-5-low · How It Works · Data refreshed daily, snapshot 2026-11-05.