gpt-6VIDEO

Tibo: Astra rewrote the inference stack to produce GPT-6.1 Sol

Relaying Tibo's account: OpenAI handed its strongest internal model Astra to top engineers to rewrite the entire inference stack, yielding GPT-6.1 Sol at about one-fifth the price, and the same model rewrote the desktop codebase serving hundreds of millions of users in 30 days.

SuSu_酥酥👅

@NFT_ChenArchived 2026.10.03Originally posted 2026.10.02

SuSu_酥酥👅 @NFT_ChenOriginally posted 2026.10.02

Open original on X

😱OpenAI 内部正在发生的事比任何发布会都更吓人！ Tibo 说：他们把最强内部模型 Astra 直接交给顶尖工程师，这些人立刻开始做以前几乎不可能的事，用 Astra 重写整套推理栈。结果就是 GPT-6.1 Sol：智能几乎接近 Astra，价格只有约 1/5。同一套内部模型还把 OpenAI 自己的交付速度打穿了。服务上亿用户的桌面端代码库，原本觉得至少要 6 个月，30 天就合并重写完，峰值每天上千 PR。先用内部模型把前沿推到极限，再用它把能力做便宜、把工程做快。这不是普通迭代，这是模型在训练下一轮自己能跑多远。 #OpenAI #Tibo #GPT #AI #Astra #AGI https://x.com/VaibhavSisinty/status/2105977760770629998/video/1

x.com/nft_chen/status/2106040170852000236Archived 2026.10.03 10:03