Atlas

Models

← All models

GLM-5.1

Zhipu AI (Z.ai) · 2026-04-07 · 753.9B parameters

GLM-5.1 is Z.ai's April 7, 2026 open-weight flagship model for agentic engineering, emphasizing coding, long-horizon planning, sustained optimization, and tool use. Its text-only MIT-licensed checkpoint contains exactly 753,864,119,552 model-weight parameters and uses a 78-layer sparse-MoE architecture with DSA over MLA, chat-template-selectable thinking, and a configured 202,752-token maximum context.

Benchmark scores

BenchmarkScore
A-Fantasia67.7
A-Fantasia Backwards Spelling78.0
A-Fantasia Chess57.0
A-Fantasia Cube Rotation68.0
AA-Briefcase Elo971
AA-LCR62.3
AA-Omniscience Index1.9
Agent Red Teaming k=100 (CAIS)75.8
Agents' Last Exam (ALE)11.5
Agents' Last Exam ALE-CLI Pass Rate12.9
Agents' Last Exam ALE-CLI Score30.8
Agents' Last Exam Full-Spectrum Pass Rate9.1
Agents' Last Exam Full-Spectrum Score21.7
Agents' Last Exam Last-Exam Pass Rate0.0
Agents' Last Exam Last-Exam Score6.2
Agents' Last Exam Near-Term Pass Rate20.1
Agents' Last Exam Near-Term Score45.6
Agents' Last Exam Overall Score28.1
Agents' Last Exam Unlicensed Full-Spectrum Pass Rate10.0
Agents' Last Exam Unlicensed Full-Spectrum Score23.0
Agents' Last Exam Unlicensed Last-Exam Pass Rate0.0
Agents' Last Exam Unlicensed Last-Exam Score6.2
Agents' Last Exam Unlicensed Near-Term Pass Rate20.1
Agents' Last Exam Unlicensed Near-Term Score45.6
Agents' Last Exam Unlicensed Overall Pass Rate12.4
Agents' Last Exam Unlicensed Overall Score29.7
AIME 2024 (avg@8)92.1
AIME 2024-202591.9
AIME 2025 (avg@8)91.7
AIME 202695.8
Loading Atlas data…