Experiences with Qwen 3.5 35B

A user shared their impressions of using the Qwen 3.5 35B language model locally, comparing it to other models like Nemotron Nano 30BA3 and GLM 4.7 Flash. The reported experience highlights how Qwen 3.5 35B surpasses previous models in several aspects.

Qwen 3.5 35B stands out for its greater overall intelligence, processing speed that does not degrade significantly with large contexts, and a superior ability to handle complex tasks. The user found that the model excels in scenarios where other models struggle, such as categorizing complex services distributed across multiple domains.

Limitations and Comparison with Other Qwen Models

Despite the advantages, the user identified some limitations, particularly in very large contexts (over 80,000 tokens), where the model may make code placement errors. It is emphasized that more advanced models would be able to handle such situations better, correctly interpreting less precise instructions.

The article presents a comparative table of the performance of different Qwen models, including Qwen 3.5 35B, Qwen 3.5 27B, and Qwen 3.5 122B, highlighting the differences in terms of speed (tokens/second), context window, and vision support. Qwen 3 Coder, a model specialized for code generation, is also mentioned.

The user wonders about the opportunity to adopt different Qwen models, evaluating the trade-off between quality and execution speed, especially in agentic development and code generation contexts, considering a workstation equipped with 48GB of VRAM.