Qwen releases open-weight driving model Qwen-Drive-1.0-4B
Around September 8, Alibaba's Qwen team published Qwen-Drive-1.0-4B, an open-weight multimodal model for autonomous driving, on Hugging Face under an Apache 2.0 license allowing free commercial use; it quickly trended on the platform.
Confirmed
- Task type is image-text-to-text, 4B parameters, Apache 2.0 license; model card, config and weights are downloadable, cross-confirmed by posts from @Qwen, @AdinaYakup and @victormustar.
- A single model covers 3D object detection, occupancy prediction and BEV perception.
- The config includes function-calling formats and vision input templates, supporting tool-calling dialogue templates.
Unconfirmed
- @FullstackSensei claims the model is fine-tuned from Qwen3.5 and that full BF16 weights are actually about 9B; this is a single-source claim not confirmed on the official model card.
Why it matters
- Unifying core driving perception tasks in one open vision-language model lowers the barrier for autonomous driving research and applications.
- The permissive Apache 2.0 license plus open weights gives small teams and researchers a directly deployable foundation model.
2026-09-08 ~ 2026-09-09 · 5 related posts
Primary sources
- [source] Qwen releases 4B autonomous-driving VLM Qwen-Drive-1.0, trending on Hugging Face — Qwen · 2026-09-08
- Qwen quietly releases Drive-1.0-4B, a 4B driving model finetuned from Qwen3.5 — FullstackSensei · 2026-09-09
- Qwen quietly ships Qwen-Drive-1.0-4B, a 4B driving model with vision and tool calling — victormustar · 2026-09-09
- [source] Qwen open-sources 4B Apache-2.0 driving VLM unifying 3D detection, occupancy and BEV — AdinaYakup · 2026-09-09
- [source] Qwen-Drive-1.0-4B model card goes live on Hugging Face with Apache 2.0 weights — AdinaYakup · 2026-09-09