Qwen3.8 Max

Qwen3.8 Max is a flagship large language model developed by Alibaba, officially launched on August 3, 2026. As the most capable model in the Qwen series to date, it features a massive 2.4 trillion parameters organized in a sparse Mixture-of-Experts (MoE) architecture. This design allows the model to activate approximately 95 billion parameters per task, balancing state-of-the-art intelligence with improved inference efficiency compared to traditional dense models of similar scale.

The model is built for "endurance" and unsupervised task completion, demonstrating significant advancements in autonomous coding, research, and long-horizon execution. In internal benchmarks, Qwen3.8 Max successfully completed a 16-day autonomous coding project, developing a self-evolving agent framework from scratch without human intervention. It natively supports multimodal inputs, including text and visual intelligence, and ranks prominently in global performance arenas, rivaling frontier models such as Claude Fable 5 and GPT-5.6 Sol.

Qwen3.8 Max supports an extensive 1 million token context window, enabling the processing of vast document repositories and complex software projects. It features a hybrid attention mechanism and support for specialized "Thinking" modes to handle highly logical reasoning and mathematical tasks. While initially accessible via the Alibaba Cloud Model Studio API, Alibaba has committed to releasing the model's weights to the open-source community, making it one of the largest open-weight models available.

The model is optimized for agentic workflows, complex reasoning, and multimodal understanding across over 100 languages. It exhibits particularly strong performance in coding benchmarks, securing high rankings in frontend development and software engineering simulations.