Compare AI models

Specifications
Parameters
16B—
Context window
32k128k
Benchmarks
HumanEval
57.3%75.7%
MATH
27.9%57.9%
MMLU
55.7%81.9%
GSM8K
72%90.9%
Overview
CompanyDeepSeekZ.ai
Release dateMay 16 2024Jun 5 2024
AccessOpen WeightClosed
Model detailsView modelView model

Frequently asked questions

GLM-4-Air leads DeepSeek V2 Lite on 4 of the 4 benchmarks they both report (HumanEval, MATH, MMLU, GSM8K). DeepSeek V2 Lite shipped 20 days before GLM-4-Air, so benchmark comparisons should account for the intervening progress.

Context windows are 32k (DeepSeek V2 Lite) vs 128k (GLM-4-Air). DeepSeek V2 Lite is open weight, while GLM-4-Air is closed.

On HumanEval, GLM-4-Air leads at 75.7% vs DeepSeek V2 Lite at 57.3%. On MATH, GLM-4-Air leads at 57.9% vs DeepSeek V2 Lite at 27.9%. On MMLU, GLM-4-Air leads at 81.9% vs DeepSeek V2 Lite at 55.7%. On GSM8K, GLM-4-Air leads at 90.9% vs DeepSeek V2 Lite at 72%.