Model Details
Gemma 4 12B
Quick answer: Gemma 4 12B is the mid-tier open-weight model in Google DeepMind's Gemma 4 family. It delivers meaningful improvements over Gemma 3 12B in reasoning, coding, and instruction following, continuing Google's pattern of advancing open-weight capability with each Gemma generation.
What Gemma 4 12B Is
Gemma 4 12B is part of Google DeepMind's fourth-generation open-weight model family, released in late 2025. The Gemma 4 generation incorporates training improvements learned from the Gemini 3 and Gemma 3n development cycles — better data quality, improved RLHF, and architectural refinements that raise the quality ceiling at each parameter count.
For teams running self-hosted inference on a single 24GB GPU, Gemma 4 12B is the direct upgrade path from Gemma 3 12B without changing hardware requirements.
Specifications
| Field | Value |
|---|---|
| Organization | Google DeepMind |
| Parameters | 12 billion |
| License | Gemma Terms of Use |
| Modality | Multimodal (text and vision) |
| Min VRAM (FP16) | ~24GB |
| HuggingFace | google/gemma-4-12b-it |
Related Models & Benchmarks
- Compare: Gemma 3 12B · Gemma 4 31B · Gemma 4 26B-A4B
- Benchmarks: MMLU-Pro
Last updated 2026-06-19.
Pricing
Open weights — self-host at no cost.
Gemma 4 12B vs Alternatives
| Model | MMLU-Pro | License |
|---|---|---|
| Gemma 4 12B | 77.2% | Gemma ToU |
| Gemma 4 26B A4B | — | — |
Check the MMLU-Pro leaderboard for current comparisons.
Frequently Asked Questions
What is Gemma 4 12B?
Google's Oct 2025 model scoring 77.2% MMLU-Pro. Gemma ToU license.Scores from Benchgen evaluations. Last updated 2026-07-24.