RunsOnMy ModelsGPUs & MacsBy VRAMAPI pricesVRAM calculatorNew

Can I run Llama Guard 4 12B locally?

Llama Guard 4 12B by meta-llama has 12B parameters. At 4-bit with an 8K context it needs about 9.1 GB of GPU memory. It fits on 65 of 65 GPUs and Macs we track. See all models that fit in 12 GB of VRAM.

9.1 GBQ4_K_M · 8K context
RTX 3060 12GBsmallest consumer GPU that fits
M1 16GBsmallest Mac that fits
10Mmax context
GQAattention type

Memory needed by quant and context length

GB of GPU memory = weights + KV cache (FP16) + runtime overhead. Weights use parameter count × bits per weight. How we calculate.

QuantQuality4K ctx8K ctx32K ctx128K ctx
2-bit (Q2_K)rough5.7 GB6.5 GB11.0 GB29.0 GB
3-bit (Q3_K_M)usable7.0 GB7.7 GB12.2 GB30.2 GB
4-bit (Q4_K_M)good, most popular8.4 GB9.1 GB13.6 GB31.6 GB
5-bit (Q5_K_M)very good9.5 GB10.2 GB14.7 GB32.7 GB
6-bit (Q6_K)near lossless10.7 GB11.5 GB16.0 GB34.0 GB
8-bit (Q8_0)effectively lossless13.4 GB14.2 GB18.7 GB36.7 GB
16-bit (F16/BF16)original weights23.9 GB24.6 GB29.1 GB47.1 GB

Which GPUs and Macs run Llama Guard 4 12B?

Best quality quant that fits at 8K context. "Tight" = within 10% of usable memory. Speed is a ceiling from memory bandwidth (real speed is typically 50–80% of it).

HardwareMemoryFits?Best quantNeedsMax tok/s (est.)
RTX 3060 12GB12 GBFitsQ5_K_M10.2 GB42details
RTX 4060 Ti 16GB16 GBFitsQ6_K11.5 GB29details
RTX 309024 GBFitsQ8_014.2 GB73details
RTX 409024 GBFitsQ8_014.2 GB79details
RTX 509032 GBFitsF1624.6 GB75details
RTX 5070 Ti16 GBFitsQ6_K11.5 GB91details
RX 7900 XTX24 GBFitsQ8_014.2 GB75details
Mac M4 Pro 48GB48 GBFitsF1624.6 GB11details
Mac M4 Max 128GB128 GBFitsF1624.6 GB23details
Mac M3 Ultra 512GB512 GBFitsF1624.6 GB34details
DGX Spark128 GBFitsF1624.6 GB11details
RTX 407012 GBFitsQ5_K_M10.2 GB59details
RTX 5060 Ti 16GB16 GBFitsQ6_K11.5 GB46details
RTX 40608 GBFitsQ2_K6.5 GB60details
Arc B58012 GBFitsQ5_K_M10.2 GB53details
RTX 3060 Ti8 GBFitsQ2_K6.5 GB100details
RTX 30708 GBFitsQ2_K6.5 GB100details
RTX 3080 10GB10 GBFitsQ3_K_M7.7 GB130details
RTX 3080 12GB12 GBFitsQ5_K_M10.2 GB107details
RTX 3090 Ti24 GBFitsQ8_014.2 GB79details
RTX 4060 Ti 8GB8 GBFitsQ2_K6.5 GB64details
RTX 4070 Super12 GBFitsQ5_K_M10.2 GB59details
RTX 4070 Ti12 GBFitsQ5_K_M10.2 GB59details
RTX 4070 Ti Super16 GBFitsQ6_K11.5 GB68details
RTX 408016 GBFitsQ6_K11.5 GB73details
RTX 4080 Super16 GBFitsQ6_K11.5 GB75details
RTX 50608 GBFitsQ2_K6.5 GB100details
RTX 507012 GBFitsQ5_K_M10.2 GB79details
RTX 508016 GBFitsQ6_K11.5 GB98details
2x RTX 30902×24 GBFitsF1624.6 GB39details
2x RTX 40902×24 GBFitsF1624.6 GB42details
2x RTX 50902×32 GBFitsF1624.6 GB75details
RTX A600048 GBFitsF1624.6 GB32details
RTX 6000 Ada48 GBFitsF1624.6 GB40details
RTX PRO 6000 Blackwell96 GBFitsF1624.6 GB75details
L40S48 GBFitsF1624.6 GB36details
A100 40GB40 GBFitsF1624.6 GB65details
A100 80GB80 GBFitsF1624.6 GB85details
H100 PCIe80 GBFitsF1624.6 GB83details
H100 SXM80 GBFitsF1624.6 GB140details
H200141 GBFitsF1624.6 GB200details
RX 76008 GBFitsQ2_K6.5 GB64details
RX 7700 XT12 GBFitsQ5_K_M10.2 GB51details
RX 7800 XT16 GBFitsQ6_K11.5 GB63details
RX 7900 GRE16 GBFitsQ6_K11.5 GB59details
RX 7900 XT20 GBFitsQ8_014.2 GB63details
RX 907016 GBFitsQ6_K11.5 GB65details
RX 9070 XT16 GBFitsQ6_K11.5 GB65details
Ryzen AI Max+ 395 128GB128 GBFitsF1624.6 GB11details
Arc A770 16GB16 GBFitsQ6_K11.5 GB57details
Mac M1 16GB16 GBFitsQ4_K_M9.1 GB9.3details
Mac M2 24GB24 GBFitsQ8_014.2 GB7.8details
Mac M3 24GB24 GBFitsQ8_014.2 GB7.8details
Mac M4 16GB16 GBFitsQ4_K_M9.1 GB16details
Mac M4 32GB32 GBFitsQ8_014.2 GB9.4details
Mac M1 Pro 32GB32 GBFitsQ8_014.2 GB16details
Mac M3 Pro 36GB36 GBFitsQ8_014.2 GB12details
Mac M4 Pro 24GB24 GBFitsQ8_014.2 GB21details
Mac M4 Pro 64GB64 GBFitsF1624.6 GB11details
Mac M1 Max 64GB64 GBFitsF1624.6 GB17details
Mac M2 Max 96GB96 GBFitsF1624.6 GB17details
Mac M3 Max 128GB128 GBFitsF1624.6 GB17details
Mac M4 Max 64GB64 GBFitsF1624.6 GB23details
Mac M2 Ultra 192GB192 GBFitsF1624.6 GB33details
Mac M3 Ultra 256GB256 GBFitsF1624.6 GB34details

Prefer an API?

Meta: Llama Guard 4 12B is on OpenRouter at $0.18 per 1M input tokens and $0.18 per 1M output tokens. Price details and history →

Questions

How much VRAM does Llama Guard 4 12B need?

About 9.1 GB at 4-bit (Q4_K_M) with an 8K context: 6.8 GB of weights (estimated), 1.5 GB of KV cache and 0.8 GB of runtime overhead. At 8-bit it needs about 14.2 GB.

What is the smallest GPU that runs Llama Guard 4 12B?

At Q4_K_M and 8K context, the smallest single consumer GPU in our table that fits it is the RTX 3060 12GB (12 GB).

Can I run Llama Guard 4 12B on a Mac?

Yes. At Q4_K_M with 8K context it fits on a Mac M1 16GB or larger, using the default macOS GPU memory limit.

Similar models

Same family or a similar parameter count, with the VRAM each needs at Q4_K_M and 8K context.

Llama 3.2 11B Vision Instruct

11B · 7.4 GB at 4-bit

Llama 3.2 11B Vision

11B · 8.0 GB at 4-bit

Llama 3.1 8B Instruct

8.0B · 6.2 GB at 4-bit

gemma 3 12b it

12B · 8.5 GB at 4-bit

GLM 4.6V Flash

10B · 6.7 GB at 4-bit

GLM 4.1V 9B Thinking

10B · 6.7 GB at 4-bit

Model facts

Hugging Face repometa-llama/Llama-Guard-4-12B
Parameters12,001,097,216 (12B)
Layers / hidden size48 / 5120
AttentionGQA; 40 heads, 8 KV heads, head dim 128
KV cache per token (FP16)192.0 KiB
Max context10,485,760 tokens
Licenseother (gated: accept terms on Hugging Face)
Released on HFApr 23, 2025
Downloads (30d) / likes171,981 / 141

Sources: Hugging Face model API · config.json (unsloth/Llama-Guard-4-12B) · GPU specs from vendor pages (list). Data fetched Nov 17, 2026. Spot an error? Report it.