Skip to content
TrustList

TrustList summary

With 72.7 billion parameters, Qwen2.5-72B-Instruct is the largest instruction-tuned model in the Qwen2.5 series from the Qwen team, and its Hugging Face page serves as the model card for downloading and running it.

What Qwen2.5 changed over Qwen2

The team released base and instruction-tuned models in this series, from 0.5 up to 72 billion parameters. Compared with Qwen2, it lists more knowledge and stronger coding and mathematics, helped by specialist expert models in those two areas. It also reports better instruction following, longer generated texts (over 8K tokens), better handling of structured data such as tables, and more reliable structured output, JSON in particular. The model is meant to cope with a wider range of system prompts, which helps role-play and chatbot set-up. More than 29 languages are supported, among them Chinese, English, French, Spanish, German, Japanese, Korean, Arabic and Vietnamese.

Architecture and context window

This repository holds a causal language model that went through pretraining and post-training. Its architecture combines transformers with RoPE, SwiGLU, RMSNorm and attention QKV bias. The card gives 70.0 billion non-embedding parameters, 80 layers, and grouped-query attention with 64 query heads and 8 key-value heads. The full context is 131,072 tokens, with generation of up to 8,192 tokens.

Running it and stretching the context

The model needs a recent release of the Hugging Face transformers library, and the card warns that versions before 4.37.0 fail with an error about the qwen2 type. A quickstart shows loading the tokenizer and model and generating a reply through the chat template. The shipped configuration handles up to 32,768 tokens; for longer input the card describes enabling YaRN rope scaling and recommends vLLM for deployment, noting that static scaling can slightly hurt performance on short texts.

Summary written by TrustList from the business's own published information, Nov 2026.

Benchmarks & AI stats

BenchmarkOfficialCommunity avg
No benchmark scores yet. Be the first to add one.

“Official” values are editor-approved and feed the ranking. “Community avg” is the mean of member submissions (shown for transparency; it never affects the ranking until an editor approves a value).

Sign in to add a benchmark score for this model.

Qwen2.5-72B launches

Request a demo or quote from Qwen2.5-72B

Protected by reCAPTCHA - Google Privacy Policy and Terms apply.

By sending, you agree we may share your request and contact details with the provider once you confirm your email.

Reviews

Write the first review of Qwen2.5-72B

Used it? Your experience helps other buyers decide.

Write a review

Questions & answers

No questions yet. Be the first to ask about Qwen2.5-72B.

Qwen2.5-72B alternatives

Similar listings buyers compare against this one.

Open-source 2B decision model from AWS's Strands Agents project that picks options and returns scores with a reliability figure

Aleph Alpha's open-weight English-German mixture-of-experts model for self-hosting, released under Apache 2.0 in October 2026

JetBrains' open 12B mixture-of-experts coding model with 2.5B parameters active per token under Apache 2.0

Liquid AI's open-weight decision model, giving yes or no, choice or score answers from text and images

Microsoft AI's real-time speech-to-text model for voice agents, transcribing 60 languages with continuous language detection

Perplexity's open-weight decision model that picks among given options for routing and checks in AI systems

Anthropic's smallest and fastest Claude 5.5 model, available on its own API and three cloud platforms

Open-weight translation model from Bilibili's Index team, covering 150 languages and following glossary and style instructions