同一提示词跑 Astra 与 Sol 的 3D 审美对比
作者用同一提示词在 Codex CLI 里跑 GPT-6 Astra 和 6.1 Sol 做 3D,Astra 花 11.52 美元,认为视觉品味仍不如 Claude。
Claude models seem to have something GPT models still haven’t quite figured out yet taste saw this fable 5.5 vs opus 5.5 output and decided to run the exact same prompt on gpt 6 astra and gpt 6.1 sol > Astra cost $11.52 > Sol cost $2.70 > both took less than 25 minutes in Codex CLI both outputs are still really good, but the difference in visual taste is hard to ignore. the little details, composition, and how intentional the whole scene feels. I’ve seen this happen across many 3D tests now, so i’m starting to think Anthropic genuinely has an edge here.
这条是引用着另一条写的:x.com/vikktorrrre/status/2105955048018588084