Source URL: https://simonwillison.net/2025/Jul/22/qwen3-235b-a22b-instruct-2507/#atom-everything
Source: Simon Willison’s Weblog
Title: Qwen/Qwen3-235B-A22B-Instruct-2507
Feedly Summary: Qwen/Qwen3-235B-A22B-Instruct-2507
Significant new model release from Qwen, published yesterday without much fanfare.
This is a follow-up to their April release of the full Qwen 3 model family, which included a Qwen3-235B-A22B model which could handle both reasoning and non-reasoning prompts (via a /no_think toggle).
The new Qwen3-235B-A22B-Instruct-2507 ditches that mechanism – this is exclusively a non-reasoning model. It looks like Qwen have new reasoning models in the pipeline.
This new model is Apache 2 licensed and comes in two official sizes: a BF16 model (437.91GB of files on Hugging Face) and an FP8 variant (220.20GB). VentureBeat estimate that the large model needs 88GB of VRAM while the smaller one should run in ~30GB.
The benchmarks on these new models look very promising. Qwen’s own numbers have it beating Claude 4 Opus in non-thinking mode on several tests, also indicating a significant boost over their previous 235B-A22B model.
I haven’t seen any independent benchmark results yet. Here’s what I got for “Generate an SVG of a pelican riding a bicycle", which I ran using the qwen3-235b-a22b-07-25:free on OpenRouter:
llm install llm-openrouter
llm -m openrouter/qwen/qwen3-235b-a22b-07-25:free \
"Generate an SVG of a pelican riding a bicycle"
Tags: ai, generative-ai, llms, llm, qwen, pelican-riding-a-bicycle, llm-release, openrouter
AI Summary and Description: Yes
Summary: The text discusses the recent release of the Qwen3-235B-A22B-Instruct-2507 model, part of the Qwen family of generative AI models. It emphasizes the shift towards non-reasoning capabilities and highlights performance benchmarks demonstrating the model’s competitive edge over other leading models in various tasks.
Detailed Description:
The provided text details a significant advancement in the generative AI landscape with the introduction of the Qwen3-235B-A22B-Instruct-2507 model. Here are the critical points regarding the model’s features, licensing, performance, and implications:
– **New Model Introduction**: The Qwen3-235B-A22B-Instruct-2507 is a follow-up model released by Qwen, following the earlier launch of the Qwen 3 family.
– **Focused Capabilities**: Unlike its predecessor, which featured a toggle for reasoning tasks, this new model is designed exclusively for non-reasoning prompts. This indicates a strategic shift in its application focus.
– **Licensing**: The model is available under Apache 2 licensing, promoting greater accessibility and collaboration within the developer community.
– **Model Sizes**: It comes in two sizes:
– BF16 model (approximately 437.91GB)
– FP8 variant (approximately 220.20GB)
– **Resource Requirements**: VentureBeat reports that the larger model requires about 88GB of VRAM, while the smaller model can operate on around 30GB.
– **Performance Benchmarks**: Initial benchmarks showcase promising results, with Qwen’s metrics indicating superiority over the Claude 4 Opus in specified non-reasoning tasks. However, independent benchmarking results are yet to be released.
– **Development Environment**: The text includes an example of how to implement the Qwen 3 model via the OpenRouter platform, illustrating practical usage for developers interested in this new release.
Overall, this release reflects ongoing innovation in the generative AI sector, with potential implications for various applications that prioritize non-reasoning tasks, affecting the strategies of professionals in AI, cloud, and software security fields. As new models emerge, understanding their capabilities and resource requirements is vital for integrating them into secure and compliant infrastructures.