Fun fact: Alper is the only person full-time training models at Sunday. It took us 3 months and 16 nodes of H100 to produce all the results you see. 2026 will be insane.
Alper Canberk
Alper Canberk9 hours ago
When training ACT-1, we treated data from diverse, long-horizon tasks in the wild as a first-class citizen. This makes generalization the default, not an exception. The capability envelope expands. More to come.
once the data bottleneck is addressed, the rest of the stack can be insanely efficient.
78.61K