international decentralized inference at 12 tok/s is unprecedented
this is the furthest anyone has gotten. Imagine pooling 500,000 gpu's worldwide for training runs that wouldn't fit in world class datacenters
couple of steps between here and there but the future is bright
1/ We published our first technical report today.
We ran a 229B model split across five consumer GPUs in five countries over the public internet and measured 12.6 tok/s interactive, 194 tok/s batched.
With cryptographic receipts on every request.
https://t.co/PrABT2u58z
two hours and I accidentally spent $2,000 on GPT 5.5 pro
you can literally rent a datacenter for $1,000/hr and be 10x as productive
safe to say every frontier lab that isn't open source is overvalued by orders of magnitude