Source URL: https://simonwillison.net/2025/Jul/30/qwen3-30b-a3b-thinking-2507/
Source: Simon Willison’s Weblog
Title: Qwen3-30B-A3B-Thinking-2507
Feedly Summary: Qwen3-30B-A3B-Thinking-2507
Yesterday was Qwen3-30B-A3B-Instruct-2507. Qwen are clearly committed to their new split between reasoning and non-reasoning models (a reversal from Qwen 3 in April), because today they released the new reasoning partner to yesterday’s model: Qwen3-30B-A3B-Thinking-2507.
I’m surprised at how poorly this reasoning mode performs at “Generate an SVG of a pelican riding a bicycle" compared to its non-reasoning partner. The reasoning trace appears to carefully consider each component and how it should be positioned… and then the final result looks like this:
I ran this using chat.qwen.ai/?model=Qwen3-30B-A3B-2507 with the "reasoning" option selected.
I also tried the "Write an HTML and JavaScript page implementing space invaders" prompt I ran against the non-reasoning model. It did a better job in that the game works:
It’s not as playable as the on I got from GLM-4.5 Air though – the invaders fire their bullets infrequently enough that the game isn’t very challenging.
This model is part of a flurry of releases from Qwen over the past two 9 days. Here’s my coverage of each of those:
Qwen3-235B-A22B-Instruct-2507 – 21st July
Qwen3-Coder-480B-A35B-Instruct – 22nd July
Qwen3-235B-A22B-Thinking-2507 – 25th July
Qwen3-30B-A3B-Instruct-2507 – 29th July
Qwen3-30B-A3B-Thinking-2507 – today
Via @Alibaba_Qwen
Tags: ai, generative-ai, llms, qwen, pelican-riding-a-bicycle, llm-reasoning, llm-release
AI Summary and Description: Yes
Summary: The text discusses the recent launch of a new reasoning model (Qwen3-30B-A3B-Thinking-2507) from Qwen that aims to improve upon previous models. It highlights the performance differences between reasoning and non-reasoning models, providing insights into the evolving landscape of generative AI. This information can be crucial for professionals focusing on AI development and security.
Detailed Description: The text highlights a recent development in the field of generative AI with the release of Qwen3-30B-A3B-Thinking-2507, a reasoning-focused AI model. Key points include:
– **Model Release**: Qwen has introduced a new model aimed at enhancing reasoning capabilities as part of a strategy to differentiate between reasoning and non-reasoning models.
– **Performance Comparison**:
– The reasoning model struggles with generating visual outputs (like SVGs), showing that prediction of image generation can be complex.
– In a comparative task, generating an HTML/JavaScript version of the game Space Invaders, the non-reasoning model performed better.
– The performance, while functional, was less engaging than earlier versions of similar models (e.g., GLM-4.5 Air).
– **Rapid Development**: Qwen has been recently active, releasing multiple models within a span of just nine days, indicating a fast-paced evolution and dedication in AI model development.
Implications for AI Security and Development:
– **Model Evaluation**: Professionals in AI and security will need to evaluate the trade-offs between reasoning and non-reasoning models when implementing AI solutions for security applications.
– **Performance Metrics**: The performance discrepancies noted when deploying these models suggest a need for rigorous benchmarking and testing.
– **Evolving AI Landscape**: Continuous updates and improvements play a crucial role in staying competitive, thus highlighting the importance of agile methodologies and rapid iteration in AI development.
In conclusion, this text is significant for professionals in AI security as it provides insights into model performance and development trends within the AI field, critical for maintaining robust security measures around generative AI applications.