Skip to main content

GLM-5.2 security benchmark rundown

GLM-5.2 (Zhipu AI) ranks 17th of 26 on BlueBench with 65.5 overall at $1.03 per task. Track scores, value against peers, strengths, weaknesses, and when to use it.

Zhipu AI · Open weight · Tested Sep 11, 2026 · Data updated Oct 2026
Methodology

Verdict

GLM-5.2 ranks 17th of 26 on BlueBench with an overall score of 65.5, in the lower half of the leaderboard and 17.7 points behind Claude Opus 5.5. At $1.03 per task it is the 12th cheapest of 26, and Claude Sonnet 5.5 scores higher (77.5) for less ($1.01). Against the 10 models priced within 2x of its cost per task, it scores 3.7 points below the peer median overall; it trails them most in threat hunting. Malware Analysis, weighted 10%, is a secondary strength (+5.5 vs the peer median).

BlueBench score
65.5
#17 of 26 models
Cost per task
$1.03
#12 cheapest of 26
Reliability
31.7
#17 of 26 models
Overall vs cost peers
−3.7
Off the cost–score frontier
All GLM-5.2 results and FAQ →

Capability by track

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

GLM-5.2 scores 65.5 overall, 17th of 26 on the capability leaderboard. Its best rank is in malware analysis (81.9, tied 8th of 26) and its worst is in threat hunting (52.4, tied 19th of 26). Among the 8 open-weight models tested it ranks 5th.

TrackWeightScoreRankPeer medianvs peers
Incident Response30%62.8#17 of 2667.0−4.2
Threat Hunting30%52.4#19 of 26 (tied)58.0−5.6
Detection Engineering30%75.9#15 of 2677.8−1.8
Malware Analysis10%81.9#8 of 26 (tied)76.4+5.5

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.

GLM-5.2 averages $1.03 per task, including repeated attempts: the 12th 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 5 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 GLM-5.2 is 3.7 points behind its price range. Reliability, 100 minus its mean score range across the benchmarks, is 31.7 (17th of 26).

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, GLM-5.2 is strongest in malware analysis (+5.5). It lags them in threat hunting (−5.6). Differences are versus the peer median score in each track; ±5 points is the threshold.

Beats its cost peers

  • Malware Analysis81.9 (+5.5)

Lags its cost peers

  • Threat Hunting52.4 (−5.6)

Nearest neighbors on the leaderboard

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

  • Claude Haiku 5.5 sits 2 places above, 3.0 points higher overall (68.5), at 58% of the cost per task ($0.59); it is 7.5 points ahead in threat hunting and 14.1 points behind in malware analysis.
  • Kimi K3 sits one place above, 1.8 points higher overall (67.3), at 2.5x the cost per task ($2.62); it is 11.9 points ahead in threat hunting and 13.9 points behind in detection engineering.
  • GPT-6 Luna sits one place below, 1.4 points lower overall (64.1), at 26% of the cost per task ($0.26); it is 8.3 points ahead in incident response and 16.6 points behind in malware analysis.
  • Gemini 3.6 Flash sits 2 places below, 2.8 points lower overall (62.7), at 2.9x the cost per task ($3.01); it is 5.2 points ahead in detection engineering and 20.3 points behind in malware analysis.
RankModelScore$/task
15Claude Haiku 5.568.5$0.59
16Kimi K367.3$2.62
17GLM-5.265.5$1.03
18GPT-6 Luna64.1$0.26
19Gemini 3.6 Flash62.7$3.01

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 (81.9, +5.5) at its price: that track beats the median of the 10 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 (+12.0) at $1.01 per task (about the same cost per task).
  • InsteadUse DeepSeek V4.1 Flash instead if threat hunting is the priority: it scores 68.5 there (vs 52.4) at $0.17 per task.
  • InsteadUse Claude Opus 5.5 instead if the highest overall score matters more than cost: it scores 83.2 (17.7 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-001 (Initial Access & Command-and-Control): 53.7 (12th of 26).
  • Its scores range from 50.8 on Simulation-002 to 75.4 on Intrusion-003, a spread of 24.6 points.
BenchmarkScoreRank
Simulation-003Impact & Exfiltration63.8#16 of 26
Simulation-002Identity & Active Directory Attacks50.8#20 of 26
Simulation-001Initial Access & Command-and-Control53.7#12 of 26
Intrusion-003AWS Cloud Intrusion75.4#13 of 26
Intrusion-002Windows Enterprise Intrusion70.6#12 of 26
Intrusion-001macOS Threat Investigation71.7#23 of 26

About this rundown

Provider
Zhipu AI · 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

Book a demo