Community Speculation on Future Qwen Open-Source Releases and Qwen 3.7 Weights
A LocalLLaMA community discussion has emerged regarding the future roadmap of Alibaba's Qwen open-source releases, specifically focusing on whether the weights for Qwen 3.7 will be publicly released. Qwen models are highly popular in the open-source AI community due to their strong performance and efficient Mixture-of-Experts (MoE) architectures. The release of Qwen 3.7 open weights would allow developers to run and fine-tune these advanced models locally. Users are currently utilizing Qwen 3.6 (such as the 35B-A3B MoE model) alongside community-tuned variants like Ornith 1.0, which is a self-improving model optimized for agentic coding. While Qwen 3.7 is accessible via OpenRouter's API, its open-source weights have not yet been officially released for local deployment.
## BACKGROUND
Qwen is a family of large language models developed by Alibaba, featuring dense models and sparse Mixture-of-Experts (MoE) models like the Qwen 3.6 35B-A3B, which uses active parameter routing to run efficiently on consumer hardware. OpenRouter is a platform that aggregates access to various LLMs via a single API, allowing developers to test models before they are open-sourced. Ornith 1.0 is a community-developed agentic coding model built using reinforcement learning on top of the Qwen architecture.