APEX-Agents-AA ↗33.70%
GLM-5.2 (Max) · APEX-Agents-AA ↗
Z.ai / MODEL EVIDENCE
GLM-5.2
55 benchmarks · 72 observations · 57 sources · released 2026-06-16
Scores at a glance
55 benchmarks · large dot: GLM-5.2; small dots: closest models where reported
EnterpriseOps-Gym-AA ↗42.73%
GLM-5.2 (Max) · EnterpriseOps-Gym-AA ↗
ITBench-AA ↗42.66%
GLM-5.2 (Max) · ITBench-AA · root-cause score ↗
Terminal-Bench Hard ↗50.76%
GLM-5.2 (Max) · Terminal-Bench Hard ↗
τ²-Bench Telecom ↗99.12%
GLM-5.2 (Max) · τ²-Bench Telecom ↗
ErdősBench ↗2.400 / 4
GLM-5.2 · ErdősBench · mean progress / 4 · API ↗
GPQA Diamond ↗85.61 %
GLM 5.2 · GPQA Diamond · Accuracy ↗
3 more reported results
91.86% · GLM-5.2 · GPQA diamond · Best score (across scorers) · % · Epoch AI · GPQA diamond ↗
87.88% · GLM-5.2 · GPQA diamond · Best score (across scorers) · % · Epoch AI · GPQA diamond ↗
71.21% · GLM-5.2 · GPQA diamond · Best score (across scorers) · % · Epoch AI · GPQA diamond ↗
GLM-5.2 · FrontierMath-Tiers-1-3-v2-Private · Best score (across scorers) · % ↗
2 more reported results
54.74% · GLM-5.2 · FrontierMath-Tiers-1-3-v2-Private · Best score (across scorers) · % · Epoch AI · FrontierMath-Tiers-1-3-v2-Private ↗
42.46% · GLM-5.2 · FrontierMath-Tiers-1-3-v2-Private · Best score (across scorers) · % · Epoch AI · FrontierMath-Tiers-1-3-v2-Private ↗
GLM-5.2 · FrontierMath-Tier-4-v2-Private · Best score (across scorers) · % ↗
GLM-5.2 · OTIS Mock AIME 2024-2025 · Best score (across scorers) · % ↗
2 more reported results
75.56% · GLM-5.2 · OTIS Mock AIME 2024-2025 · Best score (across scorers) · % · Epoch AI · OTIS Mock AIME 2024-2025 ↗
28.89% · GLM-5.2 · OTIS Mock AIME 2024-2025 · Best score (across scorers) · % · Epoch AI · OTIS Mock AIME 2024-2025 ↗
SWE-bench Verified ↗78.7%
GLM-5.2 · SWE-Bench verified · Best score (across scorers) · % ↗
1 more reported result
82.8 % · GLM 5.2 (Mini-SWE-agent) · SWE-bench Verified · Score · Vals AI · SWE-bench Verified ↗
SimpleQA Verified ↗34.2%
GLM-5.2 · SimpleQA Verified · Best score (across scorers) · % ↗
GLM-5.2 · Chess Puzzles · Best score (across scorers) · % ↗
2 more reported results
19% · GLM-5.2 · Chess Puzzles · Best score (across scorers) · % · Epoch AI · Chess Puzzles ↗
18% · GLM-5.2 · Chess Puzzles · Best score (across scorers) · % · Epoch AI · Chess Puzzles ↗
GLM-5.2 · EBR-bench · Best score (across scorers) · % ↗
GLM-5.2 · Mystery Game Puzzles · Best score (across scorers) · % ↗
2 more reported results
18% · GLM-5.2 · Mystery Game Puzzles · Best score (across scorers) · % · Epoch AI · Mystery Game Puzzles ↗
15% · GLM-5.2 · Mystery Game Puzzles · Best score (across scorers) · % · Epoch AI · Mystery Game Puzzles ↗
ARC-AGI-1 ↗77%
GLM-5.2 · ARC-AGI · Score · % ↗
SimpleBench ↗58.8%
GLM-5.2 · SimpleBench · Score (AVG@5) · % ↗
WeirdML (v2) ↗70.12%
GLM-5.2 · WeirdML · Accuracy · % ↗
1 more reported result
67.31% · GLM-5.2 · WeirdML · Accuracy · % · Epoch AI · WeirdML ↗
GLM-5.2 · APEX-Agents · Pass@1 score · % ↗
ARC-AGI-2 ↗22.78%
GLM-5.2 · ARC-AGI-2 · Score · % ↗
PostTrainBench ↗31.7%
GLM-5.2 · PostTrainBench · Average (%) · % ↗
GBAEval ↗0.02%
GLM-5.2 · GBAEval · Overall score · % ↗
CritPt ↗20.86%
GLM-5.2 · CritPt · Accuracy · % ↗
1 more reported result
3.14% · GLM-5.2 · CritPt · Accuracy · % · Epoch AI · CritPt ↗
Surface Evolver Bench ↗55.62%
GLM-5.2 · Surface Evolver Bench · Mean score · % ↗
DeepSWE v1.1 ↗43.78%
GLM-5.2 · DeepSWE · Pass@1 · % ↗
1 more reported result
36.28% · GLM-5.2 · DeepSWE · Pass@1 · % · Epoch AI · DeepSWE ↗
ProofBench ↗35%
GLM-5.2 · ProofBench · Accuracy · % ↗
FrontierCode 1.1 ↗24.5%
GLM-5.2 · FrontierCode · Main score · % ↗
LMCA ↗45.82%
GLM-5.2 · LMCA · Score · % ↗
DTBench ↗93.6%
GLM-5.2 · DTBench · Accuracy · % ↗
SciCode ↗51.16%
GLM-5.2 · SciCode · Score · % ↗
1 more reported result
36.11% · GLM-5.2 · SciCode · Score · % · Epoch AI · SciCode ↗
Text Arena (Coding) ↗1593.25 Elo
GLM-5.2 · WebDev Arena · Elo ↗
Epoch Capabilities Index (ECI) ↗151.94 ECI
GLM-5.2 · General ECI · 26 October 2026 fit ↗
Legal Research Bench ↗31.25 %
GLM 5.2 · Legal Research Bench · All-Pass ↗
LegalBench ↗84.07 %
GLM 5.2 · LegalBench · Accuracy ↗
GLM 5.2 · Harvey's Legal Agent Benchmark v1 · Accuracy ↗
Vals Finance Agent v2 ↗49.7 %
GLM 5.2 · Finance Agent v2 · Accuracy ↗
GLM 5.2 · Excel Modeling Benchmark v1 · Accuracy ↗
CorpFin v2 ↗66.12 %
GLM 5.2 · CorpFin · Accuracy ↗
MortgageTax & TaxEval v2 ↗73.34 %
GLM 5.2 · TaxEval · Accuracy ↗
MedCode ↗40.77 %
GLM 5.2 · MedCode · Accuracy ↗
MedScribe ↗83.53 %
GLM 5.2 · MedScribe · Accuracy ↗
MMLU-Pro ↗86.71 %
GLM 5.2 · MMLU Pro · Accuracy ↗
Vibe Code Bench ↗63.96 %
GLM 5.2 (OpenHands) · Vibe Code Bench · original task set ↗
ProgramBench ↗0.5 %
GLM 5.2 · ProgramBench · Accuracy ↗
SkillsBench ↗45.08 %
GLM 5.2 (OpenHands) · SkillsBench · Score ↗
Code Migration v1.1 ↗37.87 %
GLM 5.2 · Code Migration v1 · Accuracy ↗
LiveCodeBench ↗69.5 %
GLM 5.2 · LiveCodeBench · Accuracy ↗
SRE-Bench ↗0 %
GLM 5.2 · SRE Bench · Accuracy ↗
DrugDiscoveryBench ↗36.20%
GLM 5.2 (mini-SWE-agent) xhigh · DrugDiscoveryBench · Score ↗
GLM 5.2 (Mini-SWE-Agent) · SWE Atlas - Refactoring · Score ↗
GLM 5.2 (Mini-SWE-Agent) · SWE Atlas - Test Writing · Score ↗
GLM 5.2 (Mini-SWE-Agent) · Scale Labs · Score ↗
GLM 5.2 · HiL-Bench (Human-in-Loop Benchmark) · Score ↗
GLM-5.2 (max) · Kimi Code Bench 2.0 · score · Kimi K3 · official model-card comparison ↗
Vending-Bench 2 ↗8313.77833 USD
GLM-5.2 · End-of-year balance ↗
Closest models by score
Similar positions on shared benchmark leaderboards. Smaller gaps are closer.
Claude Opus 4.7 →Anthropic
21 of 31 shared benchmark comparisons
GPT-5.6 Luna →OpenAI
19 of 31 shared benchmark comparisons
GPT-5.4 →OpenAI
21 of 31 shared benchmark comparisons
How similarity is measured
Mean absolute gap in percentile positions on shared benchmark settings; lower is closer. At least 60% overlap is required, with a minimum of three when available. Independent evaluations are used when there are at least three; otherwise published report comparisons are used. Repeated model observations within one setting are averaged before ranking. Writing-style traits and composite indices are excluded. This describes similarity, not an overall quality ranking.
POST-TRAINING / SOURCE DISCLOSURE
What the GLM-5.2 / GLM-5.3 report says
Z.ai says GLM-5.3 keeps the GLM-5.2 base and scales post-training on the same long-horizon stack: SAO, asynchronous slime training and a larger mix of environments and tasks.
- Shared GLM-5.2 base checkpoint
- Single-rollout asynchronous optimization (SAO)
- Expanded long-horizon environment training
The launch report attributes the change to post-training, but its headline scores are publisher comparisons rather than a complete independent replication or an isolated optimizer ablation.
Sources
Every plotted observation links to its evaluation. Reported configurations, benchmark versions and task subsets can differ; similarity describes shared scores, not training lift.
- ARC Prize - Leaderboard ↗
- Arena leaderboard — WebDev evaluation ↗
- APEX-Agents-AA Benchmark | Artificial Analysis ↗
- EnterpriseOps-Gym-AA Benchmark | Artificial Analysis ↗
- ITBench-AA Benchmark | Artificial Analysis ↗
- SciCode Benchmark | Artificial Analysis ↗
- 𝜏²-Bench Telecom Benchmark | Artificial Analysis ↗
- Terminal-Bench Hard Benchmark | Artificial Analysis ↗
- FrontierCode Leaderboard ↗
- DTBench — Conceptual Reasoning Index ↗
- LMCA — Conceptual Reasoning Index ↗
- DeepSWE ↗
- APEX-Agents ↗
- ARC-AGI-2 ↗
- Chess Puzzles ↗
- CritPt v1 ↗
- Earthborne Rangers (EBR-bench) ↗
- FrontierMath Tier 4 (v2) ↗
- FrontierMath Tiers 1-3 (v2) ↗
- GPQA Diamond ↗
- Mystery Game Puzzles ↗
- OTIS Mock AIME 2024-2025 ↗
- PostTrainBench v1.1 ↗
- SimpleQA Verified ↗
- SWE-bench Verified ↗
- Vending Bench 2 — methodology and published results ↗
- AI benchmarks & capabilities ↗
- GLM-5.2 — GBA Eval ↗
- WeirdML ↗
- Kimi K3: Open Frontier Intelligence — official model card ↗
- DrugDiscoveryBench — evaluation methodology and results ↗
- HiL-Bench (Human-in-Loop Benchmark) — evaluation methodology and results ↗
- Scale Labs — evaluation methodology and results ↗
- SWE Atlas - Refactoring — evaluation methodology and results ↗
- SWE Atlas - Test Writing — evaluation methodology and results ↗
- AI Model Benchmarks — LM Council ↗
- ErdősBench: research mathematics, scoped progress and proof audits ↗
- Code Migration v1 — evaluation methodology and results ↗
- CorpFin — evaluation methodology and results ↗
- Excel Modeling Benchmark v1 — evaluation methodology and results ↗
- Finance Agent v2 — evaluation methodology and results ↗
- GPQA Diamond — evaluation methodology and results ↗
- Harvey's Legal Agent Benchmark v1 — evaluation methodology and results ↗
- LiveCodeBench — evaluation methodology and results ↗
- LegalBench — evaluation methodology and results ↗
- Legal Research Bench — evaluation methodology and results ↗
- MedCode — evaluation methodology and results ↗
- MedScribe — evaluation methodology and results ↗
- MMLU Pro — evaluation methodology and results ↗
- ProgramBench — evaluation methodology and results ↗
- ProofBench — evaluation methodology and results ↗
- SkillsBench — evaluation methodology and results ↗
- SRE Bench — evaluation methodology and results ↗
- SWE-bench Verified — evaluation methodology and results ↗
- TaxEval — evaluation methodology and results ↗
- Vibe Code Bench — evaluation methodology and results ↗
- Surface Evolver Bench ↗
Names used in source tables
GLM 5.2 · GLM 5.2 (Mini-SWE-Agent) · GLM 5.2 (Mini-SWE-agent) · GLM 5.2 (OpenHands) · GLM 5.2 (mini-SWE-agent) xhigh · GLM-5.2 · GLM-5.2 (Max) · GLM-5.2 (max)