Compare AI models

Specifications
Parameters
236B—
Context window
128k128k
Benchmarks
HumanEval
81.1%75.7%
MATH
53.9%57.9%
MMLU
77.8%81.9%
GSM8K
92.2%90.9%
Overview
CompanyDeepSeekZ.ai
Release dateMay 6 2024Jun 5 2024
AccessOpen WeightClosed
Model detailsView modelView model

Frequently asked questions

DeepSeek V2 and GLM-4-Air are evenly matched across the 4 benchmarks they both report (HumanEval, MATH, MMLU, GSM8K). DeepSeek V2 shipped 30 days before GLM-4-Air, so benchmark comparisons should account for the intervening progress.

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

On HumanEval, DeepSeek V2 leads at 81.1% vs GLM-4-Air at 75.7%. On MATH, GLM-4-Air leads at 57.9% vs DeepSeek V2 at 53.9%. On MMLU, GLM-4-Air leads at 81.9% vs DeepSeek V2 at 77.8%. On GSM8K, DeepSeek V2 leads at 92.2% vs GLM-4-Air at 90.9%.