GPT-5.6 Luna security benchmark rundown
GPT-5.6 Luna (OpenAI) ranks 20th of 26 on BlueBench with 62.4 overall at $0.58 per task. Track scores, value against peers, strengths, weaknesses, and when to use it.
Verdict
GPT-5.6 Luna ranks 20th of 26 on BlueBench with an overall score of 62.4, in the bottom quarter of the leaderboard and 20.8 points behind Claude Opus 5.5. At $0.58 per task it is the 6th cheapest of 26, and DeepSeek V4.1 Flash scores higher (71.5) for less ($0.17). Against the 6 models priced within 2x of its cost per task, it scores 4.6 points below the peer median overall; it trails them most in detection engineering. Malware Analysis, weighted 10%, is a secondary strength (+8.0 vs the peer median).
- BlueBench score
- 62.4
- #20 of 26 models
- Cost per task
- $0.58
- #6 cheapest of 26
- Reliability
- 33.4
- #16 of 26 models
- Overall vs cost peers
- −4.6
- 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%.
GPT-5.6 Luna scores 62.4 overall, 20th of 26 on the capability leaderboard. Its best rank is in malware analysis (79.1, 12th of 26) and its worst is in threat hunting (52.4, tied 19th of 26).
| Track | Weight | Score | Rank | Peer median | vs peers |
|---|---|---|---|---|---|
| Incident Response | 30% | 61.8 | #18 of 26 | 64.2 | −2.4 |
| Threat Hunting | 30% | 52.4 | #19 of 26 (tied) | 56.8 | −4.4 |
| Detection Engineering | 30% | 67.5 | #19 of 26 | 76.7 | −9.2 |
| Malware Analysis | 10% | 79.1 | #12 of 26 | 71.1 | +8.0 |
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.
GPT-5.6 Luna averages $0.58 per task, including repeated attempts: the 6th cheapest of 26 models with a cost figure. It is off the cost–score frontier: DeepSeek V4.1 Flash scores 71.5 at $0.17, and 3 other models are also both cheaper and at least as good. Its peer group is the 6 models priced within 2x of its cost per task. Their median overall score is 67.0, so GPT-5.6 Luna is 4.6 points behind its price range. Reliability, 100 minus its mean score range across the benchmarks, is 33.4 (16th of 26).
Cost peers: the 6 models priced within 2x of its cost per task
Strengths and weaknesses
Each track compared with the median of the 6 models priced within 2x of its cost per task. A gap of 5 points or more counts.
Against the 6 models priced within 2x of its cost per task, GPT-5.6 Luna is strongest in malware analysis (+8.0). It lags them in detection engineering (−9.2). Differences are versus the peer median score in each track; ±5 points is the threshold.
Beats its cost peers
- Malware Analysis79.1 (+8.0)
Lags its cost peers
- Detection Engineering67.5 (−9.2)
Nearest neighbors on the leaderboard
The models ranked just above and below it overall, with the largest track gaps.
- GPT-6 Luna sits 2 places above, 1.7 points higher overall (64.1), at 45% of the cost per task ($0.26); it is 9.3 points ahead in incident response and 13.8 points behind in malware analysis.
- Gemini 3.6 Flash sits one place above, 0.3 points higher overall (62.7), at 5.2x the cost per task ($3.01); it is 13.6 points ahead in detection engineering and 17.5 points behind in malware analysis.
- DeepSeek V4 Flash sits one place below, 3.8 points lower overall (58.6), at 29% of the cost per task ($0.17); it is 16.3 points behind in malware analysis.
- MiniMax M3 sits 2 places below, 6.0 points lower overall (56.4), at 1.5x the cost per task ($0.90); it is 10.7 points behind in detection engineering.
| Rank | Model | Score | $/task |
|---|---|---|---|
| 18 | GPT-6 Luna | 64.1 | $0.26 |
| 19 | Gemini 3.6 Flash | 62.7 | $3.01 |
| 20 | GPT-5.6 Luna | 62.4 | $0.58 |
| 21 | DeepSeek V4 Flash | 58.6 | $0.17 |
| 22 | MiniMax M3 | 56.4 | $0.90 |
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 (79.1, +8.0) at its price: that track beats the median of the 6 models priced within 2x of its cost per task by 5 points or more.
- InsteadUse DeepSeek V4.1 Flash instead in most cases: it scores 71.5 overall (+9.1) at $0.17 per task (30% of the cost per task).
- InsteadUse Claude Opus 5.5 instead if the highest overall score matters more than cost: it scores 83.2 (20.8 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.
- Its best standing is on Intrusion-001 (macOS Threat Investigation): 91.6 (10th of 26).
- Its scores range from 42.8 on Simulation-001 to 91.6 on Intrusion-001, a spread of 48.8 points.
| Benchmark | Score | Rank |
|---|---|---|
| Simulation-003Impact & Exfiltration | 61.8 | #20 of 26 |
| Simulation-002Identity & Active Directory Attacks | 54.0 | #19 of 26 |
| Simulation-001Initial Access & Command-and-Control | 42.8 | #22 of 26 |
| Intrusion-003AWS Cloud Intrusion | 63.8 | #20 of 26 |
| Intrusion-002Windows Enterprise Intrusion | 61.0 | #20 of 26 |
| Intrusion-001macOS Threat Investigation | 91.6 | #10 of 26 |
About this rundown
- Provider
- OpenAI · 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