Compare AI models

Specifications
Parameters
16B6B
Context window
4k32k
Benchmarks
HumanEval
45.7%9.8%
MATH
15.2%6.9%
MMLU
47.2%45.2%
GSM8K
62.2%25.9%
Overview
CompanyDeepSeekZ.ai
Release dateJan 11 2024Jun 25 2023
AccessOpen WeightOpen Weight
Model detailsView modelView model

Frequently asked questions

DeepSeek-MoE leads ChatGLM2-6B on 4 of the 4 benchmarks they both report (HumanEval, MATH, MMLU, GSM8K). ChatGLM2-6B shipped 200 days before DeepSeek-MoE, so benchmark comparisons should account for the intervening progress.

DeepSeek-MoE has 16B parameters, while ChatGLM2-6B has 6B. Context windows are 4k (DeepSeek-MoE) vs 32k (ChatGLM2-6B).

On HumanEval, DeepSeek-MoE leads at 45.7% vs ChatGLM2-6B at 9.8%. On MATH, DeepSeek-MoE leads at 15.2% vs ChatGLM2-6B at 6.9%. On MMLU, DeepSeek-MoE leads at 47.2% vs ChatGLM2-6B at 45.2%. On GSM8K, DeepSeek-MoE leads at 62.2% vs ChatGLM2-6B at 25.9%.