Alibaba's Qwen teaser 'The becoming begins' hits 13,700 likes; Qwen 4 preview testers claim 'Opus 5.5-class'

"The becoming begins." Alibaba's official Qwen account posted three words on Nov. 6 and within a day they passed about 13,700 likes and 1.47 million views. More than 1,270 replies followed, most guessing that Qwen 4 arrives today.

The post names no product. But the same day, users on X began posting results from what they called a Qwen 4 27B preview. David Y. Tan wrote that initial results look good (about 1,350 likes), then added that "as far as I can see" it is on par with Astra and Opus 5.5, "just an early testing snapshot" (about 550). Joey posted that "Qwen 4 is in my hands," showing a one-shot HTML gravity simulation with 300-plus bodies (about 760). Alok wrote that Qwen 4 27B and Flash on 24GB of VRAM would make 3090s and 4090s feel like data centers (about 590).

There is pushback. One user replied that Opus 5.5 is not beatable anytime soon, and a YouTube live stream was titled "Qwen 4 benchmarking (delayed again)." Alibaba has not announced weights, pricing or official benchmarks for the preview. On Oct. 14 Qwen said Qwen3.8-Flash-Next reached 685 output tokens per second per user on Nvidia GPUs.

Qwen also matters as the base for open models. The same day Nous Research released Arete-27B, a retrained Qwen3.8-27B (see the next item). If Qwen 4 ships with open weights, the starting point for such derivatives changes.