Compare AI models

Specifications
Parameters
7B/13B/33B/65B14.3B
Context window
2k32k
Benchmarks
HumanEval
23.7%34.2%
MMLU
63.4%62.5%
GSM8K
50.9%61.5%
Overview
CompanyMetaQwen
Release dateFeb 24 2023Mar 28 2024
AccessOpen WeightOpen Weight
Model detailsView modelView model

Frequently asked questions

Qwen1.5-MoE leads LLaMA 1 on 2 of the 3 benchmarks they both report (HumanEval, MMLU, GSM8K). LLaMA 1 shipped 398 days before Qwen1.5-MoE, so benchmark comparisons should account for the intervening progress.

LLaMA 1 has 7B/13B/33B/65B parameters, while Qwen1.5-MoE has 14.3B. Context windows are 2k (LLaMA 1) vs 32k (Qwen1.5-MoE).

On HumanEval, Qwen1.5-MoE leads at 34.2% vs LLaMA 1 at 23.7%. On MMLU, LLaMA 1 leads at 63.4% vs Qwen1.5-MoE at 62.5%. On GSM8K, Qwen1.5-MoE leads at 61.5% vs LLaMA 1 at 50.9%.