Gemma 4 12B
12 B params · 262144 context · Open weights
Released 2026-05-23
Details
API pricing
No hosted API price verified. Open-weight models can be self-hosted.
Running it locally
Memory needed for 11.96 B parameters at 8,192 tokens of context. Architecture not recorded, so a flat 1.0 GB KV allowance is used. Estimates, not measurements.
| Format | Weights | KV cache | Overhead | Total |
|---|---|---|---|---|
| Q4 K_M | 7.3 GB | 1.0 GB | 0.9 GB | 9.2 GB |
| Q6 K | 9.8 GB | 1.0 GB | 1.0 GB | 11.8 GB |
| Q8 0 | 12.7 GB | 1.0 GB | 1.1 GB | 14.8 GB |
| FP16 | 23.9 GB | 1.0 GB | 1.7 GB | 26.6 GB |
Catalogued hardware that fits at Q4
- NVIDIA GeForce RTX 5090comfortable
- NVIDIA RTX PRO 4500 Blackwellcomfortable
- NVIDIA GeForce RTX 4090comfortable
- NVIDIA RTX PRO 4000 Blackwellcomfortable
- AMD Radeon RX 7900 XTXcomfortable
- AMD Radeon RX 7900 XTcomfortable
- NVIDIA GeForce RTX 4070 Ti SUPERcomfortable
- NVIDIA GeForce RTX 4080comfortable
- NVIDIA GeForce RTX 4080 SUPERcomfortable
- NVIDIA GeForce RTX 5060 Ticomfortable
- NVIDIA GeForce RTX 5070 Ticomfortable
- NVIDIA GeForce RTX 5080comfortable
- NVIDIA RTX 2000E Ada Generationcomfortable
- NVIDIA RTX PRO 2000 Blackwellcomfortable
- AMD Radeon RX 9060 XT 16GBcomfortable
- AMD Radeon RX 9070comfortable
- AMD Radeon RX 9070 XTcomfortable
- Intel Arc B580comfortable
- NVIDIA GeForce RTX 4070 SUPERcomfortable
- NVIDIA GeForce RTX 5070comfortable
- Intel Arc B570tight
- GMKtec EVO-X3 (Ryzen AI Max+ 395)comfortable
- Beelink GTR9 Pro (Ryzen AI Max+ 395)comfortable
- OneXPlayer OneXStation i1comfortable
- Beelink GTi15 Ultra (Core Ultra 9 285H)comfortable
- Minisforum MS-02 Ultracomfortable
- Beelink SEi14 (Core Ultra 9 185H)comfortable
- Beelink SER9 Pro (Ryzen AI 9 HX 370)comfortable
- Beelink SER8 (Ryzen 7 8845HS)comfortable
- Framework Desktopcomfortable
- Apple Mac mini (M5 Pro)comfortable
- Apple Mac mini (M6)comfortable
- ASUS NUC 14 Procomfortable
- ASUS NUC 15 Procomfortable
Mini PCs reserve 8.0 GB for the operating system. “Comfortable” keeps at least 10% memory spare.
Benchmarks
No benchmarks for this model yet
We only publish results we can evidence, so there is nothing here yet and we have not filled the gap with estimates. Here is how we will test and label results.