@zephyr_z9 On the topic of Moonshot affording compute, there's a risk… I think the ultra-sparse approach is failing to handle logical reasoning and problem solving like frontier models. Made a new benchmark digging deeper:

→ View original post on X — @alexjc,