Hugging Face has launched the Gemma-4-21B-REAP model, showcasing strong performance in reasoning tasks. The model, released on April 6, demonstrates improved accuracy in these tasks, according to its developers. It is optimized for efficiency, requiring only 12GB of VRAM for limited context operations and 16GB for full context. The developers are inviting members of the MLX and GGUF communities to explore its capabilities.