Qwen3.6-Plus officially launched, with a 1 million context window, compatible with Claude Code and OpenClaw

According to 1M AI News monitoring, the Alibaba Tongyi Qianwen team has released Qwen3.6-Plus, a new generation flagship model following the Qwen3.5 series earlier this year in February. Its API is available for immediate access through Alibaba Cloud’s Bailian. The core upgrade direction focuses on Agent programming capabilities, while also strengthening multimodal perception and reasoning. By default, it supports a 1 million token context window.

In terms of programming, Qwen3.6-Plus performs outstandingly on benchmarks such as code repair, terminal operations, and automated task execution, achieving the best results across multiple long-horizon planning and tool-calling benchmarks. Its frontend development capabilities have also been improved, enabling it to handle complex projects such as 3D scenes and games. In the multimodal domain, the model shows enhancements across tasks including complex document understanding, video reasoning, and visual programming. It supports scenarios such as generating code based on interface screenshots and recreating frontend pages from design mockups.

The API has added a preserve_thinking parameter, which can retain the reasoning-thought content from prior turns in multi-turn conversations. The official recommendation is to use it for Agent tasks, saying that having a complete reasoning context helps improve decision consistency and reduce token consumption caused by repeated reasoning. Alibaba Cloud’s Bailian also supports both the OpenAI-compatible protocol and the Anthropic-compatible protocol; the latter means you can directly integrate it with Claude Code. In addition, it is compatible with programming tools such as OpenClaw, Qwen Code, Kilo Code, Cline, and OpenCode. Using Qwen Code with OAuth login provides 1,000 free calls per day.

The Qianwen team said that the focus of its recent work will shift toward the overall release of the Qwen3.6 series, and that it will also open-source smaller-scale model versions afterward.