Google Releases Gemma 4 Quantization-Aware Checkpoints for Mobile and Consumer GPUs

Google released quantization-aware training checkpoints for Gemma 4 on June 5, 2026. E2B, E4B, 12B, 26B MoE, and 31B variants come in Q4_0 GGUF and mobile compression formats, with E2B under 1GB. The checkpoints retain more quality than post-training quantization and are available under Apache 2.0 on Hugging Face. Supported runtimes include llama.cpp, Ollama, LM Studio, vLLM, SGLang, MLX, Transformers.js, and LiteRT-LM.

Visit original source