← Back to live feed1 story
The multimodal Mixture-of-Experts (MoE) system will be available for open-weight download starting tomorrow, August 26. Alibaba designed the model with 120 billion total parameters and 6 billion active parameters to showcase a next-generation framework. This release provides developers early access to the structural improvements that will underpin the upcoming Qwen4 family, marking the first public look at the new model generation.
UnslothAI is preparing day zero support for the release to facilitate immediate integration for users. The shift toward a 120B parameter scale follows the previous Qwen-3.8 27b version and signals a move toward larger capacities for open-weight multimodal models.
Sign in to suggest edits
Key sources
- SOURCE@bdsqlsz“The countdown starts now!”x.com
- SUPPORT@kimmonismus“Alibaba is giving developers early access to the architectural improvements that will underpin the upcoming model family.”x.com
- SUPPORT@jun_song“Qwen3.8-120B/51B/A6B MoE is coming out in 24 hours.”x.com
- SUPPORT@danielhanchen“working on @UnslothAI day zero support.”x.com
- SUPPORT@thezachmueller“Welcome to the era of ~120B models.”x.com
- SOURCE@alibaba_qwen“Priced at $0.40/$3 per M input/output tokens, it has shifted the Pareto frontier”x.com
- SUPPORT@kimmonismus“roughly 176B parameters of stored capacity with just 6B activated per token”x.com
- SUPPORT@wesroth“the ONLY model in its size class in the top 10”x.com