Gemma 4 12B
Google · Gemma
Google's latest Gemma generation doubles the context window to 256K tokens and — a major shift from earlier Gemma releases — ships under the Apache 2.0 license. The 12B model is one of the most actively fine-tuned coding bases on HuggingFace in 2026, with hundreds of community GGUF variants. Runs on any GPU with 10 GB VRAM.
Params
12B
License
Apache 2.0
Context
256K
Min VRAM
8 GB
Min RAM
16 GB
Tier
Standard
Run locally with Ollama
ollama run gemma4:12bOnce running, Ollama exposes an OpenAI-compatible endpoint at localhost:11434/v1.
Will it run on my hardware?
Gemma 4 12B needs about 8 GB of fast memory at Q4. Here is what in our hardware database clears that bar.
Runs at full speed · 12
Runs, but slower on shared memory · 3
Hardware links are affiliate links. We earn a small commission if you buy through them — at no extra cost to you. Disclosure
Specifications
| Parameters | 12B |
|---|---|
| Context window | 256K |
| License | Apache 2.0 |
| Min VRAM (Q4) | 8 GB |
| Min RAM | 16 GB |
| Recommended tier | Standard |
| Best for | codingchatlong documentsanalysis |
Variants on Ollama
Ollama serves the Q4_K_M quantization by default. Higher-precision tags exist for users with more memory.
Gemma 4 12B
ollama run gemma4:12bollama pull gemma4:q8_0ollama pull gemma4:fp16Exact quantization tags vary per model — run ollama show gemma4 or check the tags page for the full list.
Popularity trend (illustrative)
Illustrative adoption trend, not actual download counts. We do not publish a download number for any model because we have no honest source for it.
Similar models
Qwen3.8-27B
Alibaba · Apache 2.0
Alibaba's newest Qwen release, published August 14, 2026 under Apache 2.0. A 27B dense model that adds native image and video understanding on top of text, with a 256K context window long enough for hour-scale video or entire codebases. At Q4 it needs about 18 GB of VRAM, so it fits a single 24 GB card with room to spare. Pull it with ollama run qwen3.8:27b.
Min VRAM
18 GB
Context
256K
Run with Ollama
ollama run qwen3.8:27bMistral Nemo 12B
Mistral AI · Apache 2.0
Built in partnership with NVIDIA and trained on a broad multilingual corpus, Nemo is Mistral's best model for non-English tasks across European and Asian languages. The 128K context window makes it practical for processing long documents locally.
Min VRAM
7 GB
Context
128K
Run with Ollama
ollama run mistral-nemo:12bPhi-4 14B
Microsoft · MIT
Microsoft's research-driven Phi-4 focuses on synthetic high-quality training data, producing a 14B model that outperforms much larger models on coding and mathematical reasoning benchmarks. MIT licensed with zero usage restrictions for commercial projects.
Min VRAM
8 GB
Context
16K
Run with Ollama
ollama run phi4:14b