Alibaba Releases Qwen3.8-2.4T MoE Model on Hugging Face
Original: Qwen3.8-2.4T
Why This Matters
A 2.4T-parameter open MoE model signals Alibaba's continued push to compete at the frontier of open-weight AI.
Alibaba's Qwen team published Qwen3.8-2.4T-A95B, a Mixture-of-Experts large language model, on Hugging Face. The model features 2.4 trillion total parameters with 95 billion active parameters, and supports advanced reasoning, tool calling, and multi-turn conversation via a structured chat template.
Alibaba's Qwen team has released Qwen3.8-2.4T-A95B on Hugging Face, a large-scale Mixture-of-Experts (MoE) language model with 2.4 trillion total parameters and approximately 95 billion active parameters per forward pass. The model card reveals a detailed chat template supporting tool calling via structured XML tags, reasoning instructions, and multi-step tool use in agentic workflows. The template handles system messages, user turns, assistant turns with optional chain-of-thought reasoning enclosed in think tags, and tool response integration. The model is part of the broader Qwen3 series, continuing Alibaba's push into frontier open-weight AI models. The MoE architecture allows the model to maintain a large total parameter count while activating only a subset during inference, reducing compute costs relative to dense models of equivalent scale. No benchmark scores or licensing details are included in the available model card content, but the architecture and template design suggest targeting advanced reasoning and agentic use cases.