Search papers, labs, and topics across Lattice.
1
6
4
8
OpenAI's 120B parameter GPT-OSS model gets beat by its smaller 20B sibling on key benchmarks, challenging the assumption that bigger sparse models are always better.
