The Qwen series from Alibaba's Tongyi Lab has moved from research curiosity to a model family you actually consider for production workloads. Qwen 3.6 Plus continues that trajectory: a 1M-token context window, native bilingual training that holds up on English code tasks, and a per-token price that undercuts GPT-4-class and Claude-class APIs by a wide margin. If you've ignored the Chinese frontier labs because of access friction or fear of locking into a niche provider, the trade-offs have shifted enough that a fresh look is warranted. We ran Qwen 3.6 Plus through our internal eval harness alongside GPT and Claude. The headline isn't that it wins every benchmark — it doesn't. The headline is where it lands on the price/performance curve once you account for context length, and how that changes what's feasible in production.…