Qwen3.6 models mysteriously scrubbed from OpenRouter

The "scrubbing" of Qwen 3.6 from OpenRouter exposes the fragility of relying on third-party aggregators for bleeding-edge model access.

• –Qwen 3.6-27B is a breakthrough dense model featuring "Thinking Preservation," a technique that maintains reasoning context throughout multi-turn interactions.
• –With a 77.2% score on SWE-bench Verified, the 27B model punches way above its weight class, rivaling frontier models like Claude 4.5 Opus.
• –The removal might stem from the technical complexity of the models' hybrid architecture (linear + gated attention) which demands specialized inference optimization.
• –Developers building agentic workflows are the hardest hit, as the 35B (3B active) MoE variant was optimized specifically for low-latency tool-use tasks.