Gemma 4 31B
by Google · gemma-4 family
31B
parameters
text-generation code-generation reasoning multilingual vision tool-use math creative-writing summarization
Gemma 4 31B is the flagship of the Gemma 4 family — a 30.7B dense model that ranks #3 on Arena AI among all open models, outcompeting models with 20x more parameters. It scores 84.3% on GPQA Diamond, 89.2% on AIME 2026, and 80.0% on LiveCodeBench. At Q4 it needs about 22 GB VRAM, fitting on a RTX 3090/4090/5090 or a Mac with 24 GB+ unified memory. Released under Apache 2.0, it's one of the most permissively licensed frontier-class open models available.
Quick Start with Ollama
ollama run 31b-it-q4_K_M | Creator | |
| Parameters | 31B |
| Architecture | transformer-decoder |
| Context | 256K tokens |
| Released | Apr 2, 2026 |
| License | Apache 2.0 |
| Ollama | gemma4:31b |
Quantization Options
| Format | File Size | VRAM Required | Quality | Ollama Tag |
|---|---|---|---|---|
| Q4_K_M rec | 20 GB | 22 GB | | 31b-it-q4_K_M |
| Q8_0 | 34 GB | 38 GB | | 31b-it-q8_0 |
| F16 | 62 GB | 66 GB | | 31b-it-fp16 |
Compatible Hardware
Q4_K_M requires 22 GB VRAM
Benchmark Scores
85.2
mmlu_pro
84.3
gpqa_diamond
89.2
aime2026
80.0
livecodebench