
SenseNova Loop 9B Preview: an open model with a recursive architecture
We are used to thinking that jumping up in model quality means blindly doubling parameters: if you had 9 billion, now you need 27, along with a second GPU. The SenseNova team went against the grain and rolled out the open-source SenseNova Loop 9B Preview under the Apache 2.0 license, which outperforms 27B heavyweights on financial analysis tasks.
All the magic is hidden not in a bloated dataset, but in cyclically passing hidden states through a block of middle layers: the network literally double-checks itself before answering. However, the massive memory savings come at the cost of per-token latency.
















