【AI+9988】Alibaba releases programming model Qwen3.6-Plus, emphasizing intelligent agent programming capabilities

Alibaba (09988) On Thursday (the 2nd), Alibaba officially rolled out Qwen3.6-Plus, the next-generation large language model Qianwen. The model has been significantly upgraded in its technical architecture, with key optimizations focused on “agentic coding” and native multimodal reasoning capabilities, aiming to meet the growing enterprise demand for automated navigation and handling complex engineering tasks.

According to a technical white paper released by Alibaba Cloud, compared with the previous-generation models, Qwen3.6-Plus has achieved notable improvements in execution efficiency and reasoning stability. Its core specifications include:

A 1-million-Token context window: The model comes standard with support for an input context of 1 million tokens, allowing developers to import an entire code repository (Repository-level) or long-form technical documentation in one go. This addresses the “context switching gaps” problem that previously occurred when models had to handle large projects.

Hybrid Mixture-of-Experts (MoE) architecture optimization: It adopts a new-generation hybrid architecture to improve the model’s response speed when performing complex logic reasoning. Third-party test data shows that its generation speed is better than competing products in specific benchmark tests.

A “Perceive–Reason–Act” loop: Unlike traditional one-way generation, Qwen3.6-Plus has been deeply optimized for “capability iteration,” enabling it to complete a closed-loop workflow in a single process—from goal breakdown, to code writing, to self-testing.

Strengthened multimodal and front-end development capabilities: In multimodal application scenarios, Qwen3.6-Plus demonstrates direct transformation capabilities from visual inputs to code outputs. The model can now interpret UI screenshots, hand-drawn sketches, and product prototype diagrams, and can generate functional front-end code in real time. This technical breakthrough is seen as a key milestone in shortening product development cycles.

Enterprise-grade AI application “Wukong” has already been connected

On the day of release, Alibaba’s enterprise-grade AI application “Wukong” has already completed model integration first. Wukong member users can experience Qwen3.6-Plus in version 0.9.20 and above. Developers, meanwhile, can call the model’s API services through Alibaba Cloud’s Bai Lian platform.

According to Alibaba Cloud’s latest pricing on the Bai Lian platform, the Qwen3.6-Plus model has a minimum input cost of RMB 2 per million Tokens; Qianwen 3.6 has also been launched on Wukong, the Qianwen app, and other Alibaba AI applications and platforms. Alibaba has also previewed that it will roll out the Qwen3.6-Max version with an even larger parameter scale.

		Finance Hot Talk
	





	U.S.-Iran war risk of going on for the long haul? Is the market underestimating the global recession risk?