Claude Opus 4.8 security benchmark rundown
Claude Opus 4.8 (Anthropic) ranks 8th of 26 on BlueBench with 74.9 overall at $1.56 per task. Track scores, value against peers, strengths, weaknesses, and when to use it.
Verdict
Claude Opus 4.8 ranks 8th of 26 on BlueBench with an overall score of 74.9, in the upper half of the leaderboard and 8.3 points behind Claude Opus 5.5. At $1.56 per task it is the 14th cheapest of 26, and Claude Sonnet 5.5 scores higher (77.5) for less ($1.01). Against the 11 models priced within 2x of its cost per task, it scores 7.6 points above the peer median overall; it is strongest against them in threat hunting. Its results vary more than most across the benchmarks (reliability 17.3, 24th of 26).
- BlueBench score
- 74.9
- #8 of 26 models
- Cost per task
- $1.56
- #14 cheapest of 26
- Reliability
- 17.3
- #24 of 26 models
- Overall vs cost peers
- +7.6
- 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 4.8 scores 74.9 overall, 8th of 26 on the capability leaderboard. Its best rank is in detection engineering (85.3, 2nd of 26) and its worst is in incident response (68.4, 15th of 26).
| Track | Weight | Score | Rank | Peer median | vs peers |
|---|---|---|---|---|---|
| Incident Response | 30% | 68.4 | #15 of 26 | 69.1 | −0.7 |
| Threat Hunting | 30% | 68.6 | #4 of 26 | 56.2 | +12.4 |
| Detection Engineering | 30% | 85.3 | #2 of 26 | 77.4 | +7.9 |
| Malware Analysis | 10% | 81.9 | #8 of 26 (tied) | 77.2 | +4.7 |
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 4.8 averages $1.56 per task, including repeated attempts: the 14th cheapest of 26 models with a cost figure. It is off the cost–score frontier: Claude Sonnet 5.5 scores 77.5 at $1.01, and 1 other model is also both cheaper and at least as good. Its peer group is the 11 models priced within 2x of its cost per task. Their median overall score is 67.3, so Claude Opus 4.8 is 7.6 points ahead of its price range. Reliability, 100 minus its mean score range across the benchmarks, is 17.3 (24th of 26).
Cost peers: the 11 models priced within 2x of its cost per task
Strengths and weaknesses
Each track compared with the median of the 11 models priced within 2x of its cost per task. A gap of 5 points or more counts.
Against the 11 models priced within 2x of its cost per task, Claude Opus 4.8 is strongest in threat hunting (+12.4) and detection engineering (+7.9). Differences are versus the peer median score in each track; ±5 points is the threshold.
Beats its cost peers
- Threat Hunting68.6 (+12.4)
- Detection Engineering85.3 (+7.9)
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-5.6 Sol sits 2 places above, 2.3 points higher overall (77.2), at 1.8x the cost per task ($2.89); it is 14.2 points ahead in incident response and 5.9 points behind in threat hunting.
- GPT-6.1 Sol sits one place above, 1.8 points higher overall (76.7), at 83% of the cost per task ($1.30); it is 18.0 points ahead in incident response and 6.9 points behind in threat hunting.
- DeepSeek V4.1 Flash sits one place below, 3.4 points lower overall (71.5), at 11% of the cost per task ($0.17); it is 7.5 points behind in incident response.
- GPT-6 Sol sits 2 places below, 3.8 points lower overall (71.1), at 1.1x the cost per task ($1.74); it is 9.5 points ahead in incident response and 12.4 points behind in threat hunting.
| Rank | Model | Score | $/task |
|---|---|---|---|
| 6 | GPT-5.6 Sol | 77.2 | $2.89 |
| 7 | GPT-6.1 Sol | 76.7 | $1.30 |
| 8 | Claude Opus 4.8 | 74.9 | $1.56 |
| 9 | DeepSeek V4.1 Flash | 71.5 | $0.17 |
| 10 | GPT-6 Sol | 71.1 | $1.74 |
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 detection engineering (85.3, 2nd of 26): it is a top-3 result among all models.
- Use itUse it for threat hunting (68.6, +12.4) at its price: that track beats the median of the 11 models priced within 2x of its cost per task by 5 points or more.
- InsteadUse Claude Sonnet 5.5 instead in most cases: it scores 77.5 overall (+2.6) at $1.01 per task (65% of the cost per task).
- InsteadUse GPT-6.1 Sol instead if incident response is the priority: it scores 86.4 there (vs 68.4) at $1.30 per task.
- InsteadUse Claude Opus 5.5 instead if the highest overall score matters more than cost: it scores 83.2 (8.3 points higher) at $3.54 per task.
Notable benchmark results
Each BlueBench benchmark is a separate intrusion, real or simulated, with its own tasks and grading.
- 3rd place on Simulation-001 (Initial Access & Command-and-Control): 68.0 (3rd of 26).
- Its scores range from 68.0 on Simulation-001 to 92.9 on Intrusion-001, a spread of 24.9 points.
| Benchmark | Score | Rank |
|---|---|---|
| Simulation-003Impact & Exfiltration | 70.5 | #9 of 26 |
| Simulation-002Identity & Active Directory Attacks | 73.6 | #8 of 26 |
| Simulation-001Initial Access & Command-and-Control | 68.0 | #3 of 26 |
| Intrusion-003AWS Cloud Intrusion | 75.0 | #14 of 26 |
| Intrusion-002Windows Enterprise Intrusion | 72.6 | #9 of 26 |
| Intrusion-001macOS Threat Investigation | 92.9 | #6 of 26 |
Notes
Written by hand; they only restate the data above.
- Detection engineering is the highest of Claude Opus 4.8’s track scores (85.3, 2nd of all models), well above its incident response (68.4) and threat hunting (68.6) scores.
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