GLM-5.3 Flash (Low): benchmark results

Provider: Zhipu. Released 2026-08-20. Access: Open.

Aggregate Rating 1687 ± 24, rank #426 of 2286 rated models, from 33 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
K-Bench98.36Overall (%, Default v1 prompt)81.1
CMDB-1500 (Image)90.67Accuracy (%), fixed denominator of 300 questions (invalid an77.8
K-Bench (Therapeutic v0)97.96Overall (%, Therapeutic v0 prompt)75.8
K-Bench - Risk (Therapeutic v0)91.83Risk Score (%, Therapeutic v0 prompt)62.1
StudyArena Text - Weighted score1225Weighted score61.5
Context Arena59.1Average Score (%)61
K-Bench - Risk91.69Risk Score (%, Default v1 prompt)52.5
ARC-AGI-227.92Accuracy (%)48.2
DuelLab GameBench 2 - Reasoning Variants46.83GameBench2 normalized score (0–100)44.8
CMDB-1500 (Text)75.75Accuracy (%), fixed denominator of 1,200 questions (invalid 44.4
CAE-Bench RC7 - Fixed Problems73.55Score (0-100), 24 tasks37.5
ARC-AGI-147Accuracy (%)30.9

Interactive version: theaggregate.ai/model?slug=glm-5-3-flash-low · How It Works · Data refreshed daily, snapshot 2026-11-20.