Claude Opus 5 security benchmark rundown
Claude Opus 5 (Anthropic) ranks 24th of 26 on BlueBench with 50.5 overall at $4.44 per task. Track scores, value against peers, strengths, weaknesses, and when to use it.
Verdict
Claude Opus 5 ranks 24th of 26 on BlueBench with an overall score of 50.5, in the bottom quarter of the leaderboard and 32.7 points behind Claude Opus 5.5. At $4.44 per task it is the 23rd cheapest of 26, and Claude Opus 5.5 scores higher (83.2) for less ($3.54). Against the 8 models priced within 2x of its cost per task, it scores 23.6 points below the peer median overall; it trails them most in incident response. Malware Analysis, weighted 10%, is a secondary weak spot (−24.0 vs the peer median). Its results vary more than most across the benchmarks (reliability 25.1, 21st of 26). It refused a material share of tasks on cybersecurity grounds, and those refusals count against its scores.
- BlueBench score
- 50.5
- #24 of 26 models
- Cost per task
- $4.44
- #23 cheapest of 26
- Reliability
- 25.1
- #21 of 26 models
- Overall vs cost peers
- −23.6
- Off 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 scores 50.5 overall, 24th of 26 on the capability leaderboard. Its best rank is in detection engineering (76.3, 14th of 26) and its worst is in malware analysis (54.7, 26th of 26).
| Track | Weight | Score | Rank | Peer median | vs peers |
|---|---|---|---|---|---|
| Incident Response | 30% | 37.8 | #24 of 26 | 79.0 | −41.2 |
| Threat Hunting | 30% | 36.1 | #25 of 26 | 66.3 | −30.1 |
| Detection Engineering | 30% | 76.3 | #14 of 26 | 80.9 | −4.6 |
| Malware Analysis | 10% | 54.7 | #26 of 26 | 78.8 | −24.0 |
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 averages $4.44 per task, including repeated attempts: the 23rd cheapest of 26 models with a cost figure. It is off the cost–score frontier: Claude Opus 5.5 scores 83.2 at $3.54, and 20 other models are also both cheaper and at least as good. Its peer group is the 8 models priced within 2x of its cost per task. Their median overall score is 74.1, so Claude Opus 5 is 23.6 points behind its price range. Reliability, 100 minus its mean score range across the benchmarks, is 25.1 (21st 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 lags in incident response (−41.2), threat hunting (−30.1) and malware analysis (−24.0). Differences are versus the peer median score in each track; ±5 points is the threshold.
Beats its cost peers
No track by 5+ points
Lags its cost peers
- Incident Response37.8 (−41.2)
- Threat Hunting36.1 (−30.1)
- Malware Analysis54.7 (−24.0)
Nearest neighbors on the leaderboard
The models ranked just above and below it overall, with the largest track gaps.
- MiniMax M3 sits 2 places above, 5.9 points higher overall (56.4), at 20% of the cost per task ($0.90); it is 17.7 points ahead in threat hunting and 19.5 points behind in detection engineering.
- DeepSeek V4 Pro sits one place above, 3.1 points higher overall (53.6), at 20% of the cost per task ($0.91); it is 13.8 points ahead in incident response and 10.7 points behind in detection engineering.
- Gemini 3.1 Pro sits one place below, 5.2 points lower overall (45.3), at about the same cost per task ($4.63); it is 7.1 points ahead in threat hunting and 20.5 points behind in detection engineering.
- Claude Sonnet 5 sits 2 places below, 5.7 points lower overall (44.8), at 40% of the cost per task ($1.77); it is 32.2 points ahead in malware analysis and 12.3 points behind in threat hunting.
| Rank | Model | Score | $/task |
|---|---|---|---|
| 22 | MiniMax M3 | 56.4 | $0.90 |
| 23 | DeepSeek V4 Pro | 53.6 | $0.91 |
| 24 | Claude Opus 5 | 50.5 | $4.44 |
| 25 | Gemini 3.1 Pro | 45.3 | $4.63 |
| 26 | Claude Sonnet 5 | 44.8 | $1.77 |
Best for
Alternatives are limited to models costing at most 1.5x as much, except the overall leader and the cheaper-model pick.
- Use itBlueBench does not single out a task for it: it is in the top 3 of no track and beats its cost peers by 5+ points in none.
- InsteadUse Claude Opus 5.5 instead in most cases: it scores 83.2 overall (+32.7) at $3.54 per task (80% of the cost per task).
- CautionTest it on your own workflows before relying on it: it refused a material share of BlueBench tasks on cybersecurity grounds.
Notable benchmark results
Each BlueBench benchmark is a separate intrusion, real or simulated, with its own tasks and grading.
- Near the bottom on Simulation-003 (Impact & Exfiltration): 47.0 (25th of 26).
- Near the bottom on Intrusion-003 (AWS Cloud Intrusion): 37.8 (25th of 26).
- Near the bottom on Intrusion-002 (Windows Enterprise Intrusion): 49.8 (25th of 26).
- Near the bottom on Intrusion-001 (macOS Threat Investigation): 62.0 (25th of 26).
- Its scores range from 37.8 on Intrusion-003 to 62.0 on Intrusion-001, a spread of 24.2 points.
- 3 recorded refusals across the benchmarks; refused attempts stay in the score.
| Benchmark | Score | Rank |
|---|---|---|
| Simulation-003Impact & Exfiltration | 47.0 | #25 of 26 |
| Simulation-002Identity & Active Directory Attacks | 48.8 | #21 of 26 |
| Simulation-001Initial Access & Command-and-Control | 46.5 | #18 of 26 |
| Intrusion-003AWS Cloud Intrusion | 37.8 | #25 of 26 |
| Intrusion-002Windows Enterprise Intrusion | 49.8 | #25 of 26 |
| Intrusion-001macOS Threat Investigation | 62.0 | #25 of 26 |
Notes
Written by hand; they only restate the data above.
- Claude Opus 5 scores 76.3 in detection engineering but 37.8 in incident response and 36.1 in threat hunting. Detection rules are scored by re-running them against the evidence; incident response and threat hunting reports are graded against a hidden rubric.
About this rundown
- Provider
- Anthropic · closed weight
- Tested
- Sep 11, 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