Prediction We will have a Kimi K3 / Fable 5 equivalent intelligence running on a single RTX 5090 in less than 18 months from now

Nice presentation. The idea is also interesting about models compression, looking forward for articles about this.

We’re working on some great content at @OsmanticAI

k3 is great, 18 months is way too long from now

@TheZachMueller Way less time

@MikeBradleyAI Exciting news for the M5 Ultra specs with this prediction 🔥🤤🤖

honestly it will be within the next 18 minutes not 18 months

In 18 months, people will take loans to get a 5090

Can I work with you

That was july.....already well underway. Very accurate prediction I think.

i dont got 18 months

Within 12 months

Would be great to have Fable intelligence on a 5090, but we also hope to see inference speed improve too

The timeline for capable systems keeps shrinking. I wonder which real world bottleneck matters most now, hardware, data, or reliable reasoning?

12

Inshallah

InshAllah

@MikeBradleyAI Going to need one hell of a harness

Só precisa de 60k

This is newer than the rtx pro 6000's ( which is due in 16 m ) I need to add it to the calendar.

My take - 12 months new architecture, 10X from the current Astra-level, running on your Mac mini

@MikeBradleyAI in 18 months 5090 wont matter anymore

Thats likely the end of subscription based models. That level of capability is pretty much good enough.

@MikeBradleyAI Good talk Sir 👍

Or 5 months

That's amazing! 🙌🙌🥳

هل سمعت عن @amd & @taalas_inc ؟ 🤨 سوف نستخدم عتاد جديد كلياً بعيداً عن rtx 5090

6 months ago, Opus 4.6 was a category leader. Qwen 3.8 27b and Qwen 3.8 Flash Next are comparable. The rate of change is accelerating!

Now that's a reasonable prediction.