LLM
GA
Open weights
Gemma 4 31B
Strong, commercially usable open model on a single GPU
Release2026-03
Context131K
Max output—
Parameters31B
Local / VRAM24–32 GB
LicenseOpen weights
Overview
Gemma 4 31B is a model in the LLM category from Google DeepMind. It was released in 2026-03. The vendor is based in US/UK. Best for: Strong, commercially usable open model on a single GPU.
Hardware: Q4 ~18-20 GB; runs on a single RTX 4090.
Strengths
- High GPQA at 31B
- Image input
- Single 4090
Limits
- Gemma license prohibits use that harms Google products
- Not as strong as Chinese open models
Specifications
| Vendor | Google DeepMind |
|---|---|
| Category | LLM |
| Release | 2026-03 |
| Status | GA |
| License | Open weights |
| License name | Gemma License |
| Commercial use | Allowed |
| Parameters | 31B |
| Architecture | dense transformer |
| Context window | 131K |
| Input modalities | Text, Image |
| Output modalities | Text |
| Prompt types | Text |
| Negative prompts | No |
| Max outputs | 1 |
| Local deployment | Yes |
| VRAM | 24–32 GB |
| System RAM | 32–64 GB |
| API access | Yes |
| Pricing | Free weights |
| Confidence | High |
| Vendor country | US/UK |
| Vendor type | Frontier lab |
Pricing & hardware
Free weights
This model can be self-hosted or run locally. Approximate GPU memory: 24–32 GB. Approximate system RAM: 32–64 GB. API access is available.