WHAT NEW IN X10.30 FRI

X 上的 AI 合集档案 · 22 个合集 · 61,986 条

gpt-6POST

GPT-6 Astra 高算力安全扫描一次烧光周额度

用户跑一次 High 档安全扫描,周额度从剩 70% 直接归零还吃掉一次重置,扫描耗时一小时,吐槽 Astra 并不省 token。

MarMar Labs
@MarMarLabs2026.09.14收录原推发布于 2026.09.08
MarMar Labs @MarMarLabs原推发布于 2026.09.08
在 X 上打开原推

Hate being that guy jumping on the GPT-6 Astra usage-limit bandwagon, but yea. It's true. Ran one security scan on High/Extra High. Went from 70% of my weekly limit left to zero. Used a banked reset. Then Astra basically went "ok I'm done, found some stuff to patch." So it burned all my usage, ate a reset, and THEN handed me useful patch feedback lol. Kind of ironic. And no, I wasn't in Fast Mode. Took 1 hour for the entire scan to complete. If Astra is supposed to be more token-efficient and faster, draining usage this fast feels like going backward. I still love the model and use it every day. But power users trying to get real work done are gonna hit this wall. What @OpenAI and @thsottiaux showed here is exactly why it matters. If I could keep the model running long enough to build at that speed, it'd be incredible. Right now the limits mean shipping can actually take longer. That's my rant. Take it or leave it, but it's honest.

这条是引用着另一条写的:「OpenAI 员工称 Astra 让内部计划提前半年」

x.com/marmarlabs/status/2096999312416797032采集于 2026.09.14 15:56