Alibaba Unveils Next-Generation Flagship Qwen3.6-Max-Preview, Featuring AI Body Programming

According to Perceive Beating monitoring, the Alibaba Thousand Whys team has released Qwen3.6-Max-Preview, positioned as an early preview of the next-generation flagship model, succeeding the existing Qwen3.6-Plus. Users can directly engage in conversation at Qwen Studio (chat.qwen.ai) and then access the model through Alibaba Cloud's API under the name qwen3.6-max-preview; the interface is compatible with OpenAI's chat completions, responses specification, and Anthropic's protocol.

This version is mainly focused on agentic coding, allowing the model to write code, run it, view errors, and use tools like a programmer to complete multi-step programming tasks. The official improvements compared to the previous generation Qwen3.6-Plus are concentrated in programming: SkillsBench +9.9, SciCode +10.8, NL2Repo +5.0, Terminal-Bench 2.0 +3.8 points; world knowledge and tool invocation formats have seen improvements ranging from 2.3 to 5.3 points for three other specific enhancements. The official statement claims the highest scores in six programming benchmarks such as SWE-bench Pro, Terminal-Bench 2.0, SciCode, among others, where QwenClawBench and QwenWebBench, based on their names, appear to be proprietary evaluation sets by Thousand Whys and should be distinguished from public benchmarks.

On the API side, a new option has been added: preserve_thinking, which, when enabled, retains the thinking content from the previous rounds. The reasoning model by default only returns the "current round of thinking" once per round. In multi-round intelligent agent dialogues, the context lacks the earlier rounds of thinking, making it more likely for the model to backtrack or forget what it has already attempted when replanning. This switch addresses this gap.