Alibaba releases Qwen3.8-Flash-Next, targeting "ultimate cost efficiency"
Event progression
- Improved parameter activation efficiency and reduced training costs.
- Released the Qwen3.8-Flash-Next model, introducing new architecture and optimizer.
- Release of Qwen3.8-Flash-Next model weights with a preview of the Qwen4 architecture.
Next signal
What's to come next is the specific release and application of the Qwen4 architecture.
Sources · 3
Experiment with Qwen3.8-Flash-Next on NVIDIA GB300 NVL72 for Agentic Coding
Alibaba has released the model weights for Qwen3.8-Flash-Next, providing a preview for the upcoming Qwen4 architecture, marking further development in multimodal mixture-of-experts (MoE) models, meaning developers can explore and evaluate larger-scale AI models.
Release of Qwen3.8-Flash-Next model weights with a preview of the Qwen4 architecture.What's to come next is the specific release and application of the Qwen4 architecture.Verify Source →Alibaba’s Qwen Team Releases Qwen3.8-Flash-Next: A 125B Multimodal MoE With 6B Active Parameters Previewing the Qwen4 Architecture
Alibaba's Qwen Team has released Qwen3.8-Flash-Next, a 125B parameter multimodal MoE model, with architectural changes including Gated DeltaNet, Qwen Sparse Attention, and the Muon optimizer, indicating Alibaba's enhanced technical strength in the multimodal AI field.
Released the Qwen3.8-Flash-Next model, introducing new architecture and optimizer.The future focus can be on the further development of the Qwen4 architecture.Verify Source →Alibaba releases Qwen3.8-Flash-Next, targeting "ultimate cost efficiency"
Alibaba has released Qwen3.8-Flash-Next, significantly improving parameter activation efficiency and reducing training costs to one-ninth of competitors like OpenAI and Anthropic, exerting price pressure on them.
Improved parameter activation efficiency and reduced training costs.Pay attention to the follow-up product launch and application landing, as well as the industry's response to cost efficiency.Verify Source →