Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology

Qwen2.5-3B

Model basics

Reasoning traces

Not supported

Thinking modes

Thinking modes not supported

Context length

32K tokens

Max output length

No data

Model type

基础大模型

Release date

2024-09-18

Model file size

6GB

MoE architecture

Total params / Active params

30.0B / N/A

Knowledge cutoff

No data

Qwen2.5-3B

Open source & experience

Code license

Apache 2.0

Weights license

Tongyi Qianwen RESEARCH LICENSE AGREEMENT- 免费商用授权

GitHub repo

https://github.com/QwenLM/Qwen2.5

Hugging Face

https://huggingface.co/Qwen/Qwen2.5-3B

Live demo

No live demo

Qwen2.5-3B

Official resources

Paper

Qwen2.5-LLM: Extending the boundary of LLMs

DataLearnerAI blog

No blog post yet

Qwen2.5-3B

API details

API speed

No data

No public API pricing yet.

Qwen2.5-3B

Benchmark Results

Qwen2.5-3B currently shows benchmark results led by GSM8K (16 / 24, score 79.10), BBH (14 / 18, score 56.30), MBPP (23 / 27, score 57.10). This page also consolidates core specs, context limits, and API pricing so you can evaluate the model from benchmark results and deployment constraints together.

Thinking

All modes Normal

综合评估

4 evaluations

Benchmark / mode

Score

Rank/total

MMLU

Off

65.60

60 / 62

BBH

Off

56.30

14 / 18

MMLU Pro

Off

34.60

111 / 114

GPQA Diamond

Off

24.30

157 / 160

数学推理

2 evaluations

Benchmark / mode

Score

Rank/total

GSM8K

Off

79.10

16 / 24

MATH

Off

42.60

37 / 41

编程与软件工程

2 evaluations

Benchmark / mode

Score

Rank/total

MBPP

Off

57.10

23 / 27

HumanEval

Off

42.10

32 / 37

View benchmark analysis Compare with other models

Qwen2.5-3B

Publisher

阿里巴巴

View publisher details

Qwen2.5-3B

Model Overview

Qwen2.5-3B是阿里巴巴开源的30亿参数规模的大语言模型。属于当前业界认为的可以在移动端运行的最高尺寸的大语言模型。该模型常规版本的文件大小为6GB，可以在手机端运行。该模型与阿里巴巴开源的其它2.5代模型最大的差异是开源协议不同。它的开源协议是Qwen Research协议，该开源协议是不允许商用的。

DataLearner 官方微信

欢迎关注 DataLearner 官方微信，获得最新 AI 技术推送