WHAT NEW IN X10.30 FRI

X 上的 AI 合集档案 · 22 个合集 · 61,986 条

gemini-4IMAGE

网传 Gemini 4 Pro 泄漏跑分,作者存疑

一张据称是 Gemini 4 Pro 的泄漏 benchmark 图,若属实则碾压 GPT-6 Astra 和 Opus 5.5,但作者提醒 Google 有刷榜和针对 SVG 做 RL 的前科。

VraserX e/acc
@VraserX2026.09.29收录原推发布于 2026.09.28
网传 Gemini 4 Pro 泄漏跑分,作者存疑
VraserX e/acc @VraserX原推发布于 2026.09.28
在 X 上打开原推

These are supposedly leaked Gemini 4 Pro benchmarks, so yeah, very likely bullshit. But if they’re real, Gemini 4 Pro would absolutely destroy GPT-6 Astra and Claude Opus 5.5. I’m staying skeptical though. Google has a history of benchmark-maxxing and RL’ing specific capabilities like SVGs because people see one insane benchmark or demo and immediately assume “AGI.” Benchmarks aren’t intelligence.

x.com/vraserx/status/2104557509398651227采集于 2026.09.29 09:56