What's this about?
The community spent day two actually testing yesterday's Alibaba flagship. Ethan Mollick's informal shader-writing test finds it "solid, but not Kimi K3 level"; Chinese KOL OrangeAI calls it "a high-value K3" that holds up in his long-context assistant product.
What exactly is a "shader test"?
Mollick's personal benchmark: ask a model to write a GPU shader (code that draws visuals directly) for an evocative scene, then say "make it better". It tests code correctness plus aesthetic judgment — informal, but a consistent yardstick across models.
What's the community saying?
274 + 67 likes — modest, day-2 digestion.
- Rough consensus forming: near-frontier, priced aggressively, below Kimi K3 on creative-coding taste.
- OrangeAI's datapoint is production use (his Cola assistant with permanent memory), not benchmarks — a different and arguably harder test.
- Everyone's waiting on next week's promised open weights before the real verdict.