While the global AI community's attention was focused on Silicon Valley, Alibaba once again proved the limits of China's reasoning models to the world with its strength. On January 26, 2026, Alibaba officially released the Qwen3-Max-Thinking, the flagship reasoning model of the Qwen series. The release of this model not only marks another leap in the performance of the Qwen family but also directly challenges GPT-5.2 and Gemini3Pro, thanks to its impressive performance in 19 authoritative benchmark tests.

Two Core Innovations: Making Reasoning Smarter and More Efficient

The power of Qwen3-Max-Thinking is not only reflected in its parameter count but also in two core technological breakthroughs:

Adaptive Tool Calling Capability: This capability gives the model stronger "action ability" and has already been launched in Qwen Chat. Based on the complexity of the task, the model can autonomously determine and accurately call various external tools, enabling AI to evolve from "just talking" to "doing better."

Test-time Scaling Technology: This is a cutting-edge technology aimed at improving reasoning performance. By dynamically expanding computing resources during testing, the model can deeply break down complex logic, ensuring that every output is thoroughly thought through.

Performance Comparison: A Highlight Moment for Domestic Large Models

In multi-dimensional performance evaluations, Qwen3-Max-Thinking demonstrated qualities comparable to the world's top models. In 19 authoritative benchmark tests covering logical reasoning, mathematical abilities, programming development, and multimodal understanding, its scores were on par with leading closed-source models such as GPT-5.2 and Gemini3Pro, successfully placing it among the top-tier models globally.

Trend Tracking: The "Rapid Development" of the Qwen Family

Looking back at Alibaba's AI progress, the evolution speed of the Qwen3 series is astonishing:

September 2025: Released Qwen3-Max-Preview with a trillion-parameter scale, laying the foundation for computing power.

November 2025: The early preview version of Qwen3-Max-Thinking was unveiled, marking the beginning of exploration into reasoning models.

December 2025: Launched the full-modal large model Qwen3-Omni-Flash, achieving real-time streaming response.

January 2026: The flagship reasoning model was officially launched, marking the comprehensive maturity of reasoning capabilities.