Atlas

Models

← All models

GLM-4.5V

Zhipu AI (Z.ai) · 2025-08-11 · 107.7B parameters

GLM-4.5V is Z.ai's August 2025 open-weight vision-language model based on GLM-4.5-Air, with a language MoE marketed as 106 billion total parameters with 12 billion active, plus a vision stack. It supports image, video, long-document, chart, GUI, and grounding tasks and adds a thinking/non-thinking mode. The model is part of the GLM-V family described with GLM-4.1V-Thinking in the GLM-V technical paper.

Benchmark scores

BenchmarkScore
AA-LCR0.0
AA-Omniscience Index-43.6
AIME 202573.0
Arena Text — Business, Management & Financial Ops (No Style Control)1308
Arena Text — Chinese (No Style Control)1325
Arena Text — Creative Writing (No Style Control)1295
Arena Text — English (No Style Control)1360
Arena Text — Entertainment, Sports & Media (No Style Control)1303
Arena Text — Exclude Ties (No Style Control)1284
Arena Text — Exclude Ties (Style Controlled)1314
Arena Text — Expert (No Style Control)1356
Arena Text — Hard Prompts (No Style Control)1337
Arena Text — Hard Prompts English (No Style Control)1357
Arena Text — Instruction Following (No Style Control)1314
Arena Text — Legal & Government (No Style Control)1342
Arena Text — Life, Physical & Social Science (No Style Control)1330
Arena Text — Longer Query (No Style Control)1306
Arena Text — Medicine & Healthcare (No Style Control)1349
Arena Text — Multi-turn (No Style Control)1334
Arena Text — Non-English (No Style Control)1304
Arena Text — Non-English (Style Controlled)1323
Arena Text — Occupational Mathematical (No Style Control)1332
Arena Text — Polish (No Style Control)1309
Arena Text — Polish (Style Controlled)1326
Arena Text — Russian (No Style Control)1296
Arena Text — Software & IT Services (No Style Control)1356
Arena Text — Spanish (No Style Control)1334
Arena Text — Writing, Literature & Language (No Style Control)1305
Arena Vision — Diagram (No Style Control)1153
Arena Vision — Diagram (Style Controlled)1169
Loading Atlas data…