Claude Opus 5.5 security benchmark rundown
Claude Opus 5.5 (Anthropic) ranks 1st of 26 on BlueBench with 83.2 overall at $3.54 per task. Track scores, value against peers, strengths, weaknesses, and when to use it.
Verdict
Claude Opus 5.5 is the top-ranked model on BlueBench, with an overall score of 83.2 across 26 models. At $3.54 per task it is the 21st cheapest of 26, and it sits on the cost–score frontier: no tested model is both cheaper and higher scoring. Against the 8 models priced within 2x of its cost per task, it scores 14.1 points above the peer median overall; it is strongest against them in incident response. Malware Analysis, weighted 10%, is a secondary strength (+14.6 vs the peer median). Its results are among the most consistent across the benchmarks (reliability 56.2, 2nd of 26).
- BlueBench score
- 83.2
- #1 of 26 models
- Cost per task
- $3.54
- #21 cheapest of 26
- Reliability
- 56.2
- #2 of 26 models
- Overall vs cost peers
- +14.1
- On the cost–score frontier
Capability by track
Incident response, threat hunting, and detection engineering each carry 30% of the BlueBench score; malware analysis carries 10%.
Claude Opus 5.5 scores 83.2 overall, 1st of 26 on the capability leaderboard. Its best rank is in malware analysis (89.4, 1st of 26) and its worst is in detection engineering (83.8, 3rd of 26). It leads every model in threat hunting and malware analysis.
| Track | Weight | Score | Rank | Peer median | vs peers |
|---|---|---|---|---|---|
| Incident Response | 30% | 87.8 | #2 of 26 | 74.8 | +13.0 |
| Threat Hunting | 30% | 76.1 | #1 of 26 | 63.5 | +12.6 |
| Detection Engineering | 30% | 83.8 | #3 of 26 | 78.5 | +5.3 |
| Malware Analysis | 10% | 89.4 | #1 of 26 | 74.8 | +14.6 |
Cost, reliability, and value
Cost is USD per evaluated task including repeated attempts. Reliability is 100 minus the mean score range across the six benchmarks.
Claude Opus 5.5 averages $3.54 per task, including repeated attempts: the 21st cheapest of 26 models with a cost figure. No tested model is both cheaper and higher scoring, so it is on the cost–score frontier. Its peer group is the 8 models priced within 2x of its cost per task. Their median overall score is 69.1, so Claude Opus 5.5 is 14.1 points ahead of its price range. Reliability, 100 minus its mean score range across the benchmarks, is 56.2 (2nd of 26).
Cost peers: the 8 models priced within 2x of its cost per task
Strengths and weaknesses
Each track compared with the median of the 8 models priced within 2x of its cost per task. A gap of 5 points or more counts.
Against the 8 models priced within 2x of its cost per task, Claude Opus 5.5 is strongest in incident response (+13.0), threat hunting (+12.6), detection engineering (+5.3) and malware analysis (+14.6). Differences are versus the peer median score in each track; ±5 points is the threshold.
Beats its cost peers
- Incident Response87.8 (+13.0)
- Threat Hunting76.1 (+12.6)
- Detection Engineering83.8 (+5.3)
- Malware Analysis89.4 (+14.6)
Lags its cost peers
No track by 5+ points
Nearest neighbors on the leaderboard
The models ranked just above and below it overall, with the largest track gaps.
- GPT-6 Astra sits one place below, 2.9 points lower overall (80.3), at 2.7x the cost per task ($9.60); it is 13.3 points behind in threat hunting.
- Gemini 4 Argon sits 2 places below, 3.3 points lower overall (79.9), at 1.9x the cost per task ($6.76); it is 16.9 points behind in malware analysis.
| Rank | Model | Score | $/task |
|---|---|---|---|
| 1 | Claude Opus 5.5 | 83.2 | $3.54 |
| 2 | GPT-6 Astra | 80.3 | $9.60 |
| 3 | Gemini 4 Argon | 79.9 | $6.76 |
Best for
Alternatives are limited to models costing at most 1.5x as much, except the overall leader and the cheaper-model pick.
- Use itUse it for malware analysis (89.4, 1st of 26), threat hunting (76.1, 1st of 26), incident response (87.8, 2nd of 26) and detection engineering (83.8, 3rd of 26): each is a top-3 result among all models.
- InsteadUse Claude Sonnet 5.5 instead if cost per task matters most: it scores 77.5 (5.7 points lower) at $1.01, 29% of the cost per task.
Notable benchmark results
Each BlueBench benchmark is a separate intrusion, real or simulated, with its own tasks and grading.
- Top score on Simulation-003 (Impact & Exfiltration): 89.8 (1st of 26).
- Top score on Simulation-002 (Identity & Active Directory Attacks): 88.2 (1st of 26).
- Top score on Simulation-001 (Initial Access & Command-and-Control): 74.2 (1st of 26).
- 2nd place on Intrusion-002 (Windows Enterprise Intrusion): 80.6 (2nd of 26).
- Its scores range from 74.2 on Simulation-001 to 89.8 on Simulation-003, a spread of 15.5 points.
| Benchmark | Score | Rank |
|---|---|---|
| Simulation-003Impact & Exfiltration | 89.8 | #1 of 26 |
| Simulation-002Identity & Active Directory Attacks | 88.2 | #1 of 26 |
| Simulation-001Initial Access & Command-and-Control | 74.2 | #1 of 26 |
| Intrusion-003AWS Cloud Intrusion | 81.4 | #9 of 26 |
| Intrusion-002Windows Enterprise Intrusion | 80.6 | #2 of 26 |
| Intrusion-001macOS Threat Investigation | 89.5 | #14 of 26 |
About this rundown
- Provider
- Anthropic · closed weight
- Tested
- Sep 30, 2026 – Oct 1, 2026
- Suite
- 74 tasks across 6 benchmarks, evaluated in a common harness
- Data updated
- Oct 27, 2026
- How it is written
- Generated from the BlueBench result files with fixed rules: cost peers are models priced within 2x per task (or the nearest by cost when fewer than three are), a track is a strength or weakness at 5+ points from the peer median, and ranks share a place on ties. The text is regenerated whenever the results change.
AI for the blue team.
Run Cotool's harness in your environment to get real security work done