WHAT NEW IN X11.05 THU

X 上的 AI 合集档案 · 22 个合集 · 67,905 条

benchmarkIMAGE

Gemini 4 Pro 跑分泄露,多项指标称超 GPT-6 Astra

一份未经官方确认的 Gemini 4 Pro 规格与跑分流出,含 2M 上下文、Terminal-Bench 95.3% 等,并称在多项测试中胜过 GPT-6 Astra。

Shub
@shub04142026.09.19收录原推发布于 2026.09.18
Gemini 4 Pro 跑分泄露,多项指标称超 GPT-6 Astra
Shub @shub0414原推发布于 2026.09.18
在 X 上打开原推

Gemini 4 Pro benchmarks just leaked and if this is real, Google is coming for the AI throne. The specs look insane: → 2M token context → 256K output reportedly being tested → $2.25 / 1M input → $11.25 / 1M output → 95.3% Terminal-Bench 2.1 → 88.7% DeepSWE v1.1 → 86.8% OSWorld-2.0 → 94.7% CharXiv Reasoning And reportedly beating GPT-6 Astra on several leaked benchmarks. Nothing is officially confirmed yet but this could get very interesting.

这条是引用着另一条写的:「新 Gemini 4 Pro 基准出炉:分数大到渲染不出来」

x.com/shub0414/status/2100954711600046422采集于 2026.09.19 10:30