Qwen3.8 Max

Qwen3.8 Max is Alibaba's flagship large language model released in August 2026 as the most powerful entry in the Qwen series to date. It is a 2.4 trillion parameter model utilizing a sparse Mixture-of-Experts (MoE) architecture, where only 95 billion parameters are activated per query. This design balances frontier-level intelligence with inference efficiency, allowing the model to handle high-complexity tasks such as autonomous coding and long-horizon planning.

The model is natively multimodal, supporting inputs of text, images, and video. It features a 1-million-token context window, enabling the processing of massive code repositories, ultra-long documents, and extended video sequences. Architecturally, it integrates a hybrid attention mechanism combining Gated DeltaNet and Qwen Sparse Attention, which optimizes performance for both short-term reasoning and long-context retrieval.

Qwen3.8 Max is specifically optimized for agentic workflows. It has demonstrated the ability to operate autonomously over multi-day periods, exemplified by its performance in the OMI CLI harness where it evolved a development project from an empty directory to production-ready status. Its capabilities extend to professional domains including legal, financial, and chip design optimization, where it performs system-level planning and self-correcting execution.

In addition to the proprietary API version, Alibaba released the model weights for the 2.4T MoE checkpoint and a smaller 27B variant for local deployment. The model supports flexible reasoning control, allowing users to tune the depth of the model's thinking process via specific inference parameters to suit different task requirements.