GLM-4 (0520)

Name: GLM-4 (0520)
Author: Zhipu AI (Z.ai)

Zhipu AI (Z.ai) · 2024-05-20

GLM-4 (0520) is the May 2024 hosted GLM-4 snapshot evaluated in Zhipu AI's ChatGLM/GLM technical report. It is a closed, API-served text model; its architecture and parameter count were not disclosed separately from the GLM-4 family.

Benchmark scores

Benchmark	Score
Arena Text — Business, Management & Financial Ops (No Style Control)	1177
Arena Text — Chinese (No Style Control)	1229
Arena Text — Creative Writing (No Style Control)	1198
Arena Text — English (No Style Control)	1242
Arena Text — Entertainment, Sports & Media (No Style Control)	1185
Arena Text — Exclude Ties (No Style Control)	1124
Arena Text — Exclude Ties (Style Controlled)	1192
Arena Text — Expert (No Style Control)	1179
Arena Text — German (No Style Control)	1184
Arena Text — Hard Prompts (No Style Control)	1212
Arena Text — Hard Prompts English (No Style Control)	1219
Arena Text — Instruction Following (No Style Control)	1203
Arena Text — Korean (No Style Control)	1042
Arena Text — Legal & Government (No Style Control)	1244
Arena Text — Life, Physical & Social Science (No Style Control)	1226
Arena Text — Longer Query (No Style Control)	1206
Arena Text — Medicine & Healthcare (No Style Control)	1203
Arena Text — Multi-turn (No Style Control)	1218
Arena Text — Non-English (No Style Control)	1200
Arena Text — Non-English (Style Controlled)	1248
Arena Text — Occupational Mathematical (No Style Control)	1230
Arena Text — Russian (No Style Control)	1213
Arena Text — Software & IT Services (No Style Control)	1226
Arena Text — Writing, Literature & Language (No Style Control)	1211
LMArena Text Arena Chinese Elo	1305
LMArena Text Arena Coding Elo	1309
LMArena Text Arena Creative Writing Elo	1241
LMArena Text Arena Elo	1273
LMArena Text Arena English Elo	1288
LMArena Text Arena Expert Elo	1253