.svg){kind=spacer}
Details
- Qwen has unveiled Qwen3-Max-Preview (Instruct), a trillion-parameter large language model that stands as the company’s largest release so far.
- The model is available immediately via the Qwen Chat interface for consumers and as an endpoint in the Alibaba Cloud API.
- Benchmark tests show Max-Preview surpasses the previous flagship, Qwen3-235B-A22B-2507, by 6–14 percentage points across industry-standard suites like MMLU, GSM8K, and BIG-Bench.
- This release is an instruction-tuned preview, enabling Qwen to gather live user feedback before launching a stable version.
- While smaller Qwen3 models (14B–72B) remain open-source, Max-Preview weights are proprietary due to their size and resource demands, though Qwen may create smaller, distilled versions for on-premise use in the future.
- Pricing and usage limits for this endpoint have not been disclosed; developers can apply through the Alibaba Cloud console.
- This marks Alibaba Cloud’s first public model to exceed one trillion parameters, a significant escalation in the scale of China-based AI offerings.
Impact
This trillion-parameter launch brings Qwen closer to global heavyweights like OpenAI’s GPT-4 and Google’s Gemini Ultra, heightening the competitive pressure on domestic Chinese rivals. By providing direct API access, Alibaba Cloud positions itself as a go-to platform for enterprises wanting cutting-edge AI without turning to U.S. tech providers. If early feedback substantiates the performance claims, this could trigger broader investment in computing power and model distillation, reshaping the Chinese AI landscape.
