Skip to main content

GPT-6 Luna security benchmark rundown

GPT-6 Luna (OpenAI) ranks 18th of 26 on BlueBench with 64.1 overall at $0.26 per task. Track scores, value against peers, strengths, weaknesses, and when to use it.

OpenAI · Closed weight · Tested Sep 30, 2026 – Oct 1, 2026 · Data updated Oct 2026
Methodology

Verdict

GPT-6 Luna ranks 18th of 26 on BlueBench with an overall score of 64.1, in the lower half of the leaderboard and 19.1 points behind Claude Opus 5.5. At $0.26 per task it is the 5th cheapest of 26, and DeepSeek V4.1 Flash scores higher (71.5) for less ($0.17). Against the 3 models priced within 2x of its cost per task, it scores 6.4 points below the peer median overall; it is strongest against them in incident response and weakest in threat hunting. Malware Analysis, weighted 10%, is a secondary weak spot (−8.8 vs the peer median).

BlueBench score
64.1
#18 of 26 models
Cost per task
$0.26
#5 cheapest of 26
Reliability
40.6
#10 of 26 models
Overall vs cost peers
−6.4
Off the cost–score frontier
All GPT-6 Luna results and FAQ →

Capability by track

Incident response, threat hunting, and detection engineering each carry 30% of the BlueBench score; malware analysis carries 10%.

GPT-6 Luna scores 64.1 overall, 18th of 26 on the capability leaderboard. Its best rank is in incident response (71.1, 13th of 26) and its worst is in threat hunting (51.0, 22nd of 26).

TrackWeightScoreRankPeer medianvs peers
Incident Response30%71.1#13 of 2660.9+10.2
Threat Hunting30%51.0#22 of 2659.6−8.6
Detection Engineering30%69.9#18 of 2672.5−2.6
Malware Analysis10%65.3#21 of 2674.1−8.8

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-6 Luna averages $0.26 per task, including repeated attempts: the 5th 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 2 other models are also both cheaper and at least as good. Its peer group is the 3 models priced within 2x of its cost per task. Their median overall score is 70.5, so GPT-6 Luna is 6.4 points behind its price range. Reliability, 100 minus its mean score range across the benchmarks, is 40.6 (10th of 26).

Cost peers: the 3 models priced within 2x of its cost per task

Strengths and weaknesses

Each track compared with the median of the 3 models priced within 2x of its cost per task. A gap of 5 points or more counts.

Against the 3 models priced within 2x of its cost per task, GPT-6 Luna is strongest in incident response (+10.2). It lags them in threat hunting (−8.6) and malware analysis (−8.8). Differences are versus the peer median score in each track; ±5 points is the threshold.

Beats its cost peers

  • Incident Response71.1 (+10.2)

Lags its cost peers

  • Threat Hunting51.0 (−8.6)
  • Malware Analysis65.3 (−8.8)

Nearest neighbors on the leaderboard

The models ranked just above and below it overall, with the largest track gaps.

  • Kimi K3 sits 2 places above, 3.2 points higher overall (67.3), at 9.9x the cost per task ($2.62); it is 13.3 points ahead in threat hunting and 7.9 points behind in detection engineering.
  • GLM-5.2 sits one place above, 1.4 points higher overall (65.5), at 3.9x the cost per task ($1.03); it is 16.6 points ahead in malware analysis and 8.3 points behind in incident response.
  • Gemini 3.6 Flash sits one place below, 1.4 points lower overall (62.7), at 11.4x the cost per task ($3.01); it is 11.2 points ahead in detection engineering and 16.6 points behind in incident response.
  • GPT-5.6 Luna sits 2 places below, 1.7 points lower overall (62.4), at 2.2x the cost per task ($0.58); it is 13.8 points ahead in malware analysis and 9.3 points behind in incident response.
RankModelScore$/task
16Kimi K367.3$2.62
17GLM-5.265.5$1.03
18GPT-6 Luna64.1$0.26
19Gemini 3.6 Flash62.7$3.01
20GPT-5.6 Luna62.4$0.58

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 incident response (71.1, +10.2) at its price: that track beats the median of the 3 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 (+7.4) at $0.17 per task (65% of the cost per task).
  • InsteadUse Claude Opus 5.5 instead if the highest overall score matters more than cost: it scores 83.2 (19.1 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 Simulation-003 (Impact & Exfiltration): 68.4 (12th of 26).
  • Its scores range from 39.5 on Simulation-001 to 87.1 on Intrusion-001, a spread of 47.6 points.
BenchmarkScoreRank
Simulation-003Impact & Exfiltration68.4#12 of 26
Simulation-002Identity & Active Directory Attacks55.4#18 of 26
Simulation-001Initial Access & Command-and-Control39.5#23 of 26
Intrusion-003AWS Cloud Intrusion78.7#12 of 26
Intrusion-002Windows Enterprise Intrusion68.4#16 of 26
Intrusion-001macOS Threat Investigation87.1#16 of 26

About this rundown

Provider
OpenAI · 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

Book a demo