Alibaba Unveils Qwen3.6-Max-Preview: A New Benchmark for Agentic Coding

Alibaba has officially launched Qwen3.6-Max-Preview, the company’s most sophisticated language model to date. Released on April 20, 2026, this flagship addition to the Qwen3.6 series establishes a new performance standard, outperforming the Qwen3.6-Plus model introduced just three weeks prior. While currently in a preview phase to gather community feedback, the model has already secured top-tier rankings across six rigorous coding benchmarks, including SWE-bench Pro and Terminal-Bench 2.0. Notably, it achieved significant gains in specialized evaluation metrics, recording a 9.9-point increase on SkillsBench and a 10.8-point jump on SciCode, underscoring its proficiency in complex technical and scientific reasoning. Designed for the next generation of autonomous engineering, the model features a massive 260,000 token context window and introduces a 'preserve_thinking' capability specifically engineered to optimize multi-turn agentic workflows. Developers can integrate the model seamlessly via Alibaba Cloud’s compatible-mode endpoint, which supports both OpenAI and Anthropic API specifications. With improved instruction-following reliability and superior tool-call formatting…