Alibaba opens next-generation flagship Qwen 3.6-Max Preview, focusing on intelligent agent programming

ME News, April 20 (UTC+8). According to Beating monitoring, the Alibaba Qianwen team released Qwen3.6-Max-Preview, positioned as an early preview of the next-generation flagship model that replaces the existing Qwen3.6-Plus. Users can chat and try it directly in Qwen Studio (chat.qwen.ai); afterward, API access is opened via Alibaba Cloud Bailian under the model name qwen3.6-max-preview. The interface is compatible with OpenAI’s chat completions and responses format standards, as well as Anthropic’s protocol.

This version is mainly aimed at agentic coding, enabling the model to write code, run it, view errors, call tools, and complete multi-step programming tasks like a programmer. The official says that, compared with the prior Qwen3.6-Plus, the improvements are concentrated in programming: SkillsBench +9.9, SciCode +10.8, NL2Repo +5.0, and Terminal-Bench 2.0 +3.8. World knowledge and tool-calling formatting follow with improvements of 2.3 to 5.3 points across the other three items.

The official claims it achieved the highest scores on six programming benchmarks including SWE-bench Pro, Terminal-Bench 2.0, and SciCode. Among them, QwenClawBench and QwenWebBench appear to be Qianwen’s self-built evaluation sets based on their naming, and should be evaluated separately from public benchmarks.

On the API side, a new preserve_thinking option has also been added. When enabled, the messages will retain the thinking content from previous rounds. The reasoning model, by default, returns only the “thinking of the current round” once per turn. In multi-turn intelligent agent conversations, the context lacks the earlier steps of reasoning; when the model replans, it is prone to backtracking or forgetting what it has already tried. This switch fills that gap.

(Source: BlockBeats)