Alibaba's Qwen4-Max takes on Opus 5.5, GPT-6, and Kimi K3.1 Pro at much lower prices
First reported by The Decoder ·
The cost per AI task falls by more than half for professional users, making advanced AI capabilities more accessible.
Alibaba has launched Qwen4-Max, a new flagship large language model designed to compete with models like OpenAI's GPT-6 and Anthropic's Opus 5.5, and Kimi K3.1 Pro. Qwen4-Max features a hybrid architecture with 2.4 trillion parameters, activating 71 billion per token, and boasts a one-million-token context window. This new model aims for professional applications in industry, science, and finance, offering significant cost reductions compared to its predecessor, with an average cost per task dropping by 62% to $1.67. The model also introduces 'Live Delivery,' a feature allowing users to monitor and adjust autonomous agent tasks in real-time. Initial benchmarks show Qwen4-Max performing strongly in agent and work tasks, though it trails some competitors in specific reasoning and knowledge benchmarks. Alibaba has also demonstrated its capability in scientific research, game development, and internal infrastructure optimization.
Alibaba's Qwen4-Max leverages a novel mixture-of-experts architecture with a hybrid attention design to significantly reduce computational costs, particularly for long context windows. By compressing prior context across multiple layers before focusing on relevant details, the model drastically cuts down on the number of active parameters per token, leading to a more than 60% reduction in task costs. This architectural innovation not only makes advanced AI more affordable for professional users but also enables new applications like real-time agent task intervention through its 'Live Delivery' feature, allowing for iterative development and error correction.
The introduction of Qwen4-Max signals an intensified price war and capability race among major AI providers, pushing the boundaries of what is achievable at lower operational expenses. While Qwen4-Max demonstrates strong performance in specialized professional tasks and autonomous agent work, its mixed results on broader benchmarks suggest a strategic focus on specific industry needs rather than universal dominance. The open-sourcing of smaller Qwen4 models is also anticipated, which could further democratize access to advanced AI technologies and spur broader ecosystem development.
AI-written summary. May contain errors.