DeepSeek V4 Pro security benchmark rundown
DeepSeek V4 Pro (DeepSeek) ranks 23rd of 26 on BlueBench with 53.6 overall at $0.91 per task. Track scores, value against peers, strengths, weaknesses, and when to use it.
Verdict
DeepSeek V4 Pro ranks 23rd of 26 on BlueBench with an overall score of 53.6, in the bottom quarter of the leaderboard and 29.6 points behind Claude Opus 5.5. At $0.91 per task it is the 9th cheapest of 26, and DeepSeek V4.1 Flash scores higher (71.5) for less ($0.17). Against the 10 models priced within 2x of its cost per task, it scores 15.6 points below the peer median overall; it trails them most in threat hunting. Malware Analysis, weighted 10%, is a secondary weak spot (−20.5 vs the peer median). Its results vary more than most across the benchmarks (reliability 31.1, 18th of 26).
- BlueBench score
- 53.6
- #23 of 26 models
- Cost per task
- $0.91
- #9 cheapest of 26
- Reliability
- 31.1
- #18 of 26 models
- Overall vs cost peers
- −15.6
- 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%.
DeepSeek V4 Pro scores 53.6 overall, 23rd of 26 on the capability leaderboard. Its best rank is in detection engineering (65.6, 21st of 26) and its worst is in threat hunting (41.3, 24th of 26). Among the 8 open-weight models tested it ranks 8th.
| Track | Weight | Score | Rank | Peer median | vs peers |
|---|---|---|---|---|---|
| Incident Response | 30% | 51.6 | #23 of 26 | 67.0 | −15.3 |
| Threat Hunting | 30% | 41.3 | #24 of 26 | 58.0 | −16.7 |
| Detection Engineering | 30% | 65.6 | #21 of 26 | 77.8 | −12.2 |
| Malware Analysis | 10% | 60.0 | #24 of 26 | 80.5 | −20.5 |
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.
DeepSeek V4 Pro averages $0.91 per task, including repeated attempts: the 9th 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 7 other models are also both cheaper and at least as good. Its peer group is the 10 models priced within 2x of its cost per task. Their median overall score is 69.2, so DeepSeek V4 Pro is 15.6 points behind its price range. Reliability, 100 minus its mean score range across the benchmarks, is 31.1 (18th of 26).
Cost peers: the 10 models priced within 2x of its cost per task
Strengths and weaknesses
Each track compared with the median of the 10 models priced within 2x of its cost per task. A gap of 5 points or more counts.
Against the 10 models priced within 2x of its cost per task, DeepSeek V4 Pro lags in threat hunting (−16.7), incident response (−15.3), detection engineering (−12.2) and malware analysis (−20.5). 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
- Threat Hunting41.3 (−16.7)
- Incident Response51.6 (−15.3)
- Detection Engineering65.6 (−12.2)
- Malware Analysis60.0 (−20.5)
Nearest neighbors on the leaderboard
The models ranked just above and below it overall, with the largest track gaps.
- DeepSeek V4 Flash sits 2 places above, 5.0 points higher overall (58.6), at 19% of the cost per task ($0.17); it is 10.6 points ahead in threat hunting.
- MiniMax M3 sits one place above, 2.8 points higher overall (56.4), at about the same cost per task ($0.90); it is 12.5 points ahead in threat hunting and 8.8 points behind in detection engineering.
- Claude Opus 5 sits one place below, 3.1 points lower overall (50.5), at 4.9x the cost per task ($4.44); it is 10.7 points ahead in detection engineering and 13.8 points behind in incident response.
- Gemini 3.1 Pro sits 2 places below, 8.3 points lower overall (45.3), at 5.1x the cost per task ($4.63); it is 19.0 points behind in incident response.
| Rank | Model | Score | $/task |
|---|---|---|---|
| 21 | DeepSeek V4 Flash | 58.6 | $0.17 |
| 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 |
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 DeepSeek V4.1 Flash instead in most cases: it scores 71.5 overall (+17.9) at $0.17 per task (19% of the cost per task).
- InsteadUse GPT-6.1 Sol instead if consistent results matter: its reliability is 52.9 (vs 31.1) with an overall score of 76.7 at $1.30 per task.
- InsteadUse Claude Opus 5.5 instead if the highest overall score matters more than cost: it scores 83.2 (29.6 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.
- Near the bottom on Simulation-002 (Identity & Active Directory Attacks): 40.9 (24th of 26).
- Near the bottom on Simulation-001 (Initial Access & Command-and-Control): 23.6 (26th of 26).
- Its scores range from 23.6 on Simulation-001 to 90.2 on Intrusion-001, a spread of 66.6 points.
| Benchmark | Score | Rank |
|---|---|---|
| Simulation-003Impact & Exfiltration | 56.5 | #22 of 26 |
| Simulation-002Identity & Active Directory Attacks | 40.9 | #24 of 26 |
| Simulation-001Initial Access & Command-and-Control | 23.6 | #26 of 26 |
| Intrusion-003AWS Cloud Intrusion | 60.0 | #21 of 26 |
| Intrusion-002Windows Enterprise Intrusion | 60.3 | #21 of 26 |
| Intrusion-001macOS Threat Investigation | 90.2 | #12 of 26 |
About this rundown
- Provider
- DeepSeek · open 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