gemini-4POST

Gemini 4 Argon vs Opus 5.5 on reproducing an exponential-time hypothesis proof

With the same prompt, Opus 5.5 took 23 minutes and 1% of weekly quota to reproduce the proof, while Gemini 4 Argon took about 9 minutes with a more concise core recursion but one wrong conclusion.

Aurora

@MingTian685474Archived 15 hours agoOriginally posted 2026.10.27

Aurora @MingTian685474Originally posted 2026.10.27

Open original on X

OpenAI 和 Anthropic 同一天宣布自家 AI 推翻了指数时间假设 Anthropic README 里说，把核心思路压成几行提示，Opus 5.5 就能重新推出证明试了一下，Opus 5.5 耗时 23 分钟，消耗 1% 周额度做出来了，Gemini 4 Argon 用同一个提示词，核心递推证明对了，其中一步还比 Opus 5.5 简洁，而且就用了差不多 9 分钟，但是多了一个错误结论不过 Anthropic 是在 OpenAI 发出后 5 小时才发布的，看起来应该是被迫发布了，而且实际上那所谓几行提示是把最难的构造都给了😂好像也不能据此看出什么问题

x.com/mingtian685474/status/2115003489726386680Archived 2026.10.28 10:02