Compare AI models

Specifications
Parameters
7B/13B/33B/65B—
Context window
2k128k
Benchmarks
HumanEval
23.7%75.7%
MATH
10.6%57.9%
MMLU
63.4%81.9%
GSM8K
50.9%90.9%
Overview
CompanyMetaZ.ai
Release dateFeb 24 2023Jun 5 2024
AccessOpen WeightClosed
Model detailsView modelView model

Frequently asked questions

GLM-4-Air leads LLaMA 1 on 4 of the 4 benchmarks they both report (HumanEval, MATH, MMLU, GSM8K). LLaMA 1 shipped 467 days before GLM-4-Air, so benchmark comparisons should account for the intervening progress.

Context windows are 2k (LLaMA 1) vs 128k (GLM-4-Air). LLaMA 1 is open weight, while GLM-4-Air is closed.

On HumanEval, GLM-4-Air leads at 75.7% vs LLaMA 1 at 23.7%. On MATH, GLM-4-Air leads at 57.9% vs LLaMA 1 at 10.6%. On MMLU, GLM-4-Air leads at 81.9% vs LLaMA 1 at 63.4%. On GSM8K, GLM-4-Air leads at 90.9% vs LLaMA 1 at 50.9%.