Google DeepMind solicits feedback for next Gemma models

The "Gemma 4 era" marks a strategic pivot toward agent-first open models, but the community is already pushing for more compute-heavy and efficient variants. High demand persists for a 124B MoE model to rival top-tier proprietary models in reasoning and complex multi-step task handling. Developers are calling for improved native tool-calling and function-calling reliability to support autonomous, proactive local agents. Additionally, 1-bit quantization and extreme context windows beyond 256K are top priorities for high-efficiency, memory-constrained edge environments, as the shift to the Apache 2.0 license has accelerated enterprise adoption.