Qwen Unveils Qwen3-Max: Its Most Advanced Open Language Model Yet

Details

• On September 28, 2025, Alibaba’s Qwen team publicly introduced Qwen3-Max, the flagship model of its third-generation series.
• The launch announcement includes a single link to supporting documentation, where details like parameter count, context window, and licensing are expected to be found.
• Qwen3-Max builds on the legacy of Qwen2-72B, which garnered top marks on benchmarks like MMLU, GSM8K, and HumanEval after its Apache 2.0 release in 2024.
• The "Max" tier has historically denoted Qwen’s largest scale model, typically featuring broader context length, multilingual capabilities, and robust coding skills.
• Engineers describe this as the initial release, hinting at upcoming variants including instruction-tuned, multimodal, and domain-specific models.
• Early users have immediate access to model weights, with an Alibaba Cloud deployment expected in the coming weeks.

Impact

Qwen3-Max's debut turns up the competitive heat on rivals like Meta's anticipated Llama 4 and Mistral’s upcoming MOE models, with open weights improving transparency. Free access to a powerful Chinese LLM may lower AI barriers for local startups and bolster China's technology self-sufficiency amid US export controls. The annual cadence of major releases positions Qwen as a long-term open-source challenger in the global model landscape.