Looks like we have a new top-tier open-source Qwen model—and I’ve got to say, they did an amazing job with this release. In this video I fully test Qwen 3.6 Plus, a new agentic coding model with a massive 1 million token context window, built for real-world tasks. The big focus here is repo-level coding, terminal tasks, automation workflows, and long-horizon planning with tool use, while also improving multimodal reasoning across images, documents, and even video. Coding-wise, this model is exceptional and I’m definitely adding it to my workflow. I put it through real builds using the Kilo CLI, including a browser-based Mac OS clone with multiple apps and real settings toggles (light/dark, display tweaks), a first-person flight HTML game, and even a surprisingly strong Minecraft-style clone with block breaking, textures, water, a cave system, ores, and lava damage. Frontend output is also very strong—sometimes comparable to Opus—especially on landing pages, SVG art/animations, and a TikTok-style mobile clone with scrolling and interactions. There are tradeoffs: when it generates long projects, it can be sluggish because it reasons for a while. But benchmark-wise it’s competing at a very high level (SWE/terminal-style tasks and multimodal benchmarks), and pricing is honestly reasonable for what you’re getting. I also show the easiest ways to try it (chat, API, OpenRouter free, and Kilo’s free API), plus why its upcoming smaller open-source variants make this drop even more interesting.