NCRI推测GPT-6 Astra 6.2基于与6.2相同的底层模型,并执行两次循环。因价格下降且层数限制,排除增加循环次数可能,认为6.2可能存在欠训练或为更大模型。
NCRI / serial reasoning depth for Astra 6.2 is a higher than I expected
so if it's the same underlying model as Astra 6 then it might be more loopy
but my default assumption is that they are both based on the same model and loop back two times
why?
another pass over the recurrent block would increase cost, but prices fell from Astra 6 to 6.2. And the 3 and 4 loop config for Astra 6.2 would execute > 240 layers, which would contradict Jakub's statement.
now, what options are still open?
- Astra 6 was undertrained and didn't yet utilize the loopies effectively (<-- my favorite)
- Astra 6.2 is a larger model
- I'm regarded and all of this is wrong
I decided against writing an article and revealing how I come to these new estimates. I want to do more analysis on the coming models. my latest central estimate is that GPT-6 Astra is: - 4.2T@120B - around 112 layers - loops 50% of its layers once I also have evidence that GPT-6 Sol and Luna, and Opus 5.5 are NOT looped language models. The leaks about "Astra Minor" suggest a smaller Sol sized looped model coming, that probably slides right into the $20-25 price point.在 X 查看被引用的帖子
来源:Lisan al Gaib · x.com