gpt-6POST
GPT-6 Astra High 登顶 PPTBench
Astra High 以 77.34 分领跑 PPTBench，但语义、渲染和小细节还原仍普遍失败。
KATE
@kate_osita_2026.09.30收录原推发布于 2026.09.29
KATE @kate_osita_原推发布于 2026.09.29
GPT-6 Astra High leads the PPTBench board at 77.34, but the score matters less than the gap it exposes. Faithful visual reconstruction still fails across: - semantics - rendering - small details The benchmark is useful because it shows exactly where visual agents need to improve.
这条是引用着另一条写的：x.com/2016491434120294400/status/210494177043381…
x.com/kate_osita_/status/2104952784013668517采集于 2026.09.30 10:24