AI Lab
Official Qwen announcements — model releases, product launches and research, each one summarized with every source covering it, by AIBriefs. RSS
Qwen-Drive-1.0-4B combines driving-scene understanding with motion planning, built on Qwen3.5-4B with added 3D perception and trajectory components. Developed with Huazhong University of Science and Technology, it's released under Apache 2.0 with code, weights, and demo data.
The open-weight multimodal MoE activates only 6B parameters per token, with a 262K native context window extensible to 1M via YaRN. QwenCloud API pricing is $0.16/1M input and $0.47/1M output tokens.
A Reddit user asks for the cheapest hardware to run Qwen 3.8 27B at ~50 tokens/s, currently owning an RTX 5070 Ti. The thread collects community recommendations for budget GPU setups.