Atlas

Models

← All models

GLM-5.2

Zhipu AI (Z.ai) · 2026-06-16 · 753.3B parameters

GLM-5.2 is Z.ai's open-weight flagship text model for long-horizon work, released June 16, 2026, with a 1,048,576-token context window and multiple thinking-effort levels. Its released 78-layer MoE checkpoint contains exactly 753,329,921,024 model parameters and uses DeepSeek Sparse Attention over MLA with IndexShare, which reuses the same indexer across every four sparse-attention layers. Z.ai releases it under MIT without regional restrictions.

Benchmark scores

BenchmarkScore
A-Fantasia68.3
A-Fantasia (Valid Attempts with Retries)68.1
A-Fantasia Backwards Spelling77.0
A-Fantasia Chess56.0
A-Fantasia Chess (Valid Attempts with Retries)56.0
A-Fantasia Cube (Valid Attempts with Retries)72.0
A-Fantasia Cube Rotation72.0
A-Fantasia Spelling (Valid Attempts with Retries)76.3
AA-Briefcase Elo1260
AA-Briefcase v1.1 Elo1227
AA-LCR71.3
AA-LCR v1.142.3
AA-Omniscience Index4.4
AdvancedIF77.4
AdvancedIF (rubric mean score)94.4
Agent Red Teaming k=100 (CAIS)71.0
Agents' Last Exam (ALE)20.4
Agents' Last Exam ALE-CLI Pass Rate23.8
Agents' Last Exam ALE-CLI Score44.2
Agents' Last Exam Full-Spectrum Pass Rate18.2
Agents' Last Exam Full-Spectrum Score37.0
Agents' Last Exam Last-Exam Pass Rate2.6
Agents' Last Exam Last-Exam Score12.9
Agents' Last Exam Near-Term Pass Rate32.8
Agents' Last Exam Near-Term Score60.3
Agents' Last Exam Overall Score41.1
Agents' Last Exam Unlicensed Full-Spectrum Pass Rate18.0
Agents' Last Exam Unlicensed Full-Spectrum Score36.8
Agents' Last Exam Unlicensed Last-Exam Pass Rate3.3
Agents' Last Exam Unlicensed Last-Exam Score13.0
Loading Atlas data…