Gemma 4 31B Dense
#45 in Open-Source LLMsgoogle · v4 · 31b dense · since 2026-04-02 · 3× · last seen Jun 30, 2026
11
Momentum
Gemma 4 31B Dense is a large language model from Google with all parameters fully active. It is designed for local high-performance applications on workstations and delivers maximum quality by utilizing all parameters.
Momentum trend
04.04.03.07.
Features
| Context Window | 256,000 tokens (256K); Proportional RoPE for long-context optimization |
| Model Size (Parameters) | 30.7B active parameters (Dense; total incl. embeddings: 33B); all parameters active at every inference step |
| Memory Requirement | BF16/FP16 (full precision): ~64–71 GB VRAM; FP8: ~32 GB VRAM (H100); Q4 quantization (local): ~18–20 GB VRAM; Recommended for comfortable local operation: 24 GB VRAM (e.g., RTX 3090/4090) |