Atlas

Models

← All models

GLM 5 Turbo

Zhipu AI (Z.ai) · 2026-03-15

GLM-5-Turbo is Z.ai's March 2026 language model optimized for OpenClaw agent workflows. It has a 200K-token context window and up to 128K output tokens, with optimizations for tool invocation, command following, scheduled and persistent tasks, and long-chain execution. It is available through Z.ai's API rather than as a public weight release.

Benchmark scores

BenchmarkScore
AA-LCR60.7
AA-LCR v1.171.7
AA-Omniscience Index-16.4
ALE-Bench1313
Artificial Analysis Omniscience Accuracy28.4
Artificial Analysis Omniscience Hallucination Rate62.6
Artificial Analysis Progress Index v1277
BullshitBench v2 Leaderboard31.0
Capability147.4
CritPt (avg@5)0.3
Design Arena Overall Code Categories1260
GPQA Diamond (Artificial Analysis avg@5)84.7
Humanity's Last Exam (Text-Only)27.8
IFBench (Artificial Analysis avg@5)73.2
ObviousBench95.1
SciCode (Artificial Analysis avg@3)43.6
SM Bench52.0
SM Bench — Adversarial (Hostile Logic)77.6
SM Bench — Ambiguous Interpretation13.7
SM Bench — Anti-hallucination18.9
SM Bench — Creative Writing (Mature Themes)55.2
SM Bench — EQ Boundaries69.1
SM Bench — NSFW (Explicit)35.8
SM Bench — NSFW (System Prompt)57.1
SM Bench — Overfit70.5
SnakeBench Average Apples7.7
SnakeBench Best Apples17.0
SnakeBench Rating26.3
SnakeBench Total Apples92.0
SnakeBench Win Rate75.0
Loading Atlas data…