Alibaba's Qwen 3.8 Max open-weights model challenges proprietary AI on coding and reasoning benchmarks.

Alibaba has introduced Qwen 3.8 Max, a frontier open-weights multimodal AI model designed to challenge leading proprietary systems from OpenAI and Anthropic. The model features a 1-million-token context window and is priced significantly lower than competitors, costing $0.0248 compared to $0.253 for Opus 5. Qwen 3.8 Max excels in agentic workflows, demonstrating the ability to operate autonomously for extended periods, including a 16-day autonomous coding test where it resolved 172 issues and completed 127 pull requests. In benchmark evaluations, Qwen 3.8 Max leads rivals on the ERQA embodied reasoning test and the PerceptionBench visual perception test. It also achieves a 56.2% score on the challenging Humanity's Last Exam benchmark when using tools. The model's multimodal capabilities enable visual coding, frontend web development, and native 3D modeling in Blender. Additionally, it supports a 100-hour video memory feature utilizing a three-level hierarchical structure to analyze long-form video content. Alibaba plans to release the model weights, alongside smaller variants like Qwen 3.6 27B and 35B.

Alibaba's Qwen 3.8 Max is an open-weights multimodal AI model featuring a 1-million-token context window. The model is priced at $0.0248, making it significantly cheaper than proprietary rivals like Opus 5 and GPT-5.6 Sol.

Qwen 3.8 Max demonstrated 16 days of autonomous coding, resolving 172 issues and committing 265 times. It outperforms competitors on the ERQA embodied reasoning and PerceptionBench visual perception benchmarks.

The model supports native 3D modeling in Blender and automated frontend web development from visual prototypes. It scored 56.2% on the Humanity's Last Exam benchmark when equipped with external tools.