Alibaba Unveils Qwen3-Max with Trillion Parameters

Alibaba Group has unveiled its latest large language model, Qwen3-Max-Preview, featuring over one trillion parameters. This model is now accessible through Alibaba Cloud API and OpenRouter, offering significant improvements in reasoning, instruction following, and multilingual support. The model supports a context window of 262,144 tokens, making it suitable for complex reasoning and coding tasks.

Qwen3-Max-Preview is designed to handle extensive inputs and outputs, with a maximum input of 258,048 tokens and a maximum output of 32,768 tokens. It includes context caching to optimize performance during extended sessions. Despite its capabilities, the model is not open-source, requiring developers to access it through paid APIs.

The pricing for Qwen3-Max-Preview is tiered based on token usage, starting at $0.861 per million input tokens and $3.441 per million output tokens for smaller tasks, scaling up for larger workloads. This structure aims to balance cost-efficiency with the model's extensive capabilities.

The release of Qwen3-Max-Preview marks a significant step for Alibaba in the AI landscape, showcasing its commitment to advancing large-scale AI models.