Activefrontierllm Open Source

Qwen1.5 72B

by Alibaba· Released February 2024· Cutoff September 2023

Qwen1.5 72B is a large language model from Alibaba's Qwen series, featuring a mixture-of-experts architecture for efficient scaling. It excels in multilingual tasks, particularly Chinese and English, and offers strong performance across a wide range of benchmarks. As the largest model in the Qwen1.5 family, it provides state-of-the-art capabilities for complex reasoning and generation.

Official Site API Docs 🤗 Hugging Face 📄 Research Paper

Input cost

Free (open source)

Output cost

Free (open source)

Context window

32768 tokens

Max output

8192 tokens

Modalities

text

Parameters

72B

License

Apache-2.0

Capabilities

Function CallingCode GenerationMultilingual SupportLong Context UnderstandingStreamingJSON Mode

Best For

Complex multilingual tasks requiring deep reasoning and long context understanding.

Strengths

Strong performance on multilingual benchmarks, especially Chinese and English
Large 32K context window for handling lengthy documents
Efficient MoE architecture balances performance and compute
Open-source with permissive license allows customization

Limitations

Requires significant computational resources for inference
May not match specialized models in narrow domains
Limited to text-only input; no vision or multimodal support
Knowledge cutoff may not cover very recent events

Use Cases

Multilingual customer support chatbots

Code generation and debugging assistance

Long-form document summarization and analysis

Content creation in Chinese and English

Complex reasoning tasks like math and logic

Translation and cross-lingual information retrieval

Research and experimentation in NLP

Improvements Over Previous Model

Introduced MoE architecture for better efficiency compared to dense Qwen 72B
Increased context window from 8K to 32K tokens
Improved multilingual performance, especially in non-English languages
Enhanced function calling and tool use capabilities
Better alignment with human preferences through advanced RLHF

Back to all models

Improvements Over Previous Model

Introduced MoE architecture for better efficiency compared to dense Qwen 72B

Increased context window from 8K to 32K tokens

Improved multilingual performance, especially in non-English languages

Enhanced function calling and tool use capabilities

Better alignment with human preferences through advanced RLHF