Tweet by u/hackerllama Could be copium, but I would love to see Gemma 4.1 there with unified audio input for all model sizes perhaps even up to 120B, much improved tool calling (even with the latest template there are still bugs), higher precision QAT from the start and improved general performance
