Qwen3.6-35B-A3B: Agentic coding power, now open to all (qwen.ai)

Article summary

The Qwen team has released Qwen3.6-35B-A3B, a new model in the Qwen 3.6 series, with open weights available. This model is intended as a superior replacement for the previous Qwen3.5-27B model. The new model delivers outstanding agentic coding performance, rivaling larger dense models. The release of this model is seen as a continuation of the Qwen team's efforts to provide open weights for their models.

Main themes

• Qwen model release
• Open weights
• Model performance
• Agentic coding
• Compute requirements

What commenters say

• The new Qwen3.6-35B-A3B model is a significant improvement over the previous Qwen3.5-27B model, but its performance may not surpass that of larger dense models in all benchmarks.
• The release of open weights for the Qwen3.6 series is a positive development, allowing for greater freedom and flexibility in using the models.
• Running large models like Qwen3.6-35B-A3B requires significant computational resources, but some users are able to run them on their own hardware, while others may need to rely on third-party providers.
• The cost of running large models locally can be prohibitively expensive for individual users, but may be within the budget of small companies or organizations.
• The Qwen team's decision to release open weights for their models is seen as a continuation of their efforts to provide accessible and flexible AI solutions.
• Some users are skeptical about the benefits of running large models locally, citing the potential for cloud-based solutions to be more cost-effective and efficient.
• The development of more full-featured software for local AI deployment is needed to make it more accessible and user-friendly for a wider range of users.
• The ability to run large models locally can provide benefits such as increased privacy and control, as well as the ability to refine models for specific use cases.