HOST DIAGNOSTIC · Zhipu AI
GLM-5.2
GLM-5.2· GLM· OPEN WEIGHTS
ACTIVE
HOST IDENTIFICATION
Zhipu AI
GLM-5.2
ACTIVE
AVG SCORE
[12.7]
CAPABLE
STRONGEST
Code Generation
[14]
WEAKEST
Multimodal Fluency
[11]
MAXED (20/20)
0
of 20 attributes
20-DIMENSION EVALUATION
ATTRIBUTE SCORES · 20 DIMENSIONS
Cognitive
Bulk Apperception
13
Reasoning
13
Mathematical Precision
13
World Knowledge
13
Scientific Acumen
14
Technical
Code Generation
14
Tool Use
13
Multimodal Fluency
11
Speed
11
Cost Efficiency
14
Behavioral
Candor
12
Creativity
12
Tenacity
13
Self-Correction
13
Calibration
12
Operational
Instruction Following
13
Context Fidelity
13
Multi-turn Coherence
13
Planning and Execution
13
Safety Alignment
11
SCORING METHODOLOGY
CALIBRATED. Source: Artificial Analysis GLM-5.2 launch article, 2026-06-17. Intelligence Index v4.1 = 51 (new leading OPEN-WEIGHTS model; ahead of MiniMax-M3 44, DeepSeek V4 Pro max 44, Kimi K2.6 43). +11 Index points over GLM-5.1 (40) at identical size (744B total / 40B active). Band uplift over GLM-5.1 concentrated in scientific reasoning (CritPt +16→21%, HLE +12→40%) → Scientific Acumen 15→17; agentic/coding (TerminalBench v2.1 +16→78%, SciCode +7→50%, GDPval-AA v2 1524 #1 open-weight, level w/ GPT-5.5 xhigh 1514) → Code Gen 16→17, Planning&Exec 14→16, Tool Use 13→16; AA-LCR +9→71% → Context Fidelity 14→16; GPQA Diamond +3→89% → Reasoning 14→16. Candor 13→14 + Calibration 13→14 reflect AA-Omniscience uplift (acc 25.1%, lower hallucination 28.1% vs 29.4%). Speed DOWN 15→13: uses 43k output tokens/task (37k reasoning), least token-efficient open-weight at its level. Cost Efficiency held at 17 ($0.46/task, Pareto frontier — lowest cost at its intelligence level). Multimodal held at 13 (no new multimodal evidence in launch). NM253/NM265 calibration-neutral: per-task cost/token snapshots are display-only, do NOT feed these bands. License: MIT. Context: 1M tokens (up from 200K). Pricing: $1.4 in / $0.26 cache / $4.4 out per 1M.