Gemma 4 12B Unified IT
Gemma 4 12B Unified IT by Google DeepMind (11.95B parameters). At Q4_K_M with an 8K context it needs about 8.0 GiB of memory. A 12B that feeds images and audio straight into the language model (no separate encoder), with a 256K context; it fits a 12 GB card at 6-bit.
Last updated: 2026-10-13
Model facts
Run it
ollama run gemma4:12b
Memory needed
Weights + KV cache + runtime overhead, in GiB. Lower quantization = less memory, slightly lower quality.
| Quantization | 4K context | 8K context | 32K context |
|---|---|---|---|
| Q4_K_M | 7.9 GiB | 8.0 GiB | 8.1 GiB |
| Q5_K_M | 9.2 GiB | 9.2 GiB | 9.4 GiB |
| Q6_K | 10.5 GiB | 10.5 GiB | 10.7 GiB |
| Q8_0 | 13.3 GiB | 13.3 GiB | 13.5 GiB |
| FP16 | 24.2 GiB | 24.2 GiB | 24.4 GiB |
Which hardware runs it
Best quantization that fits each device with a 8K context. Estimates, not benchmarks. See the methodology.
Frequently asked questions
How much VRAM does Gemma 4 12B Unified IT need?
With an 8K context, about 8.0 GiB at Q4_K_M and 13.3 GiB at Q8_0 (weights + KV cache + runtime overhead). Longer contexts need more memory for the KV cache.
Can Gemma 4 12B Unified IT run on an GeForce RTX 4090?
Yes. Gemma 4 12B Unified IT fits on the GeForce RTX 4090 (24 GB) at Q8_0 with an 8K context, using about 13.3 GiB, at an estimated 56 tokens/s.
Can I run Gemma 4 12B Unified IT on a Mac?
Yes. Gemma 4 12B Unified IT fits on the M3 Ultra (96 GB) at FP16 with an 8K context, using about 24.2 GiB, at an estimated 22 tokens/s.
Can Gemma 4 12B Unified IT be used commercially?
Gemma 4 12B Unified IT is released under Apache-2.0. The licence allows commercial use (read its conditions).
Sources
- https://ai.google.dev/gemma/docs/releases
- https://ai.google.dev/gemma/docs/gemma_4_license
- https://huggingface.co/google/gemma-4-12B-it
- https://huggingface.co/google/gemma-4-12B-it/resolve/main/config.json
- https://huggingface.co/api/models/google/gemma-4-12B-it?expand[]=safetensors
- https://ollama.com/library/gemma4/tags