Llama 4 models

2 Llama 4 models from Meta on Hugging Face, from 108.6B to 401.6B parameters, published by Meta in BF16, with a 10M-token context. The smallest official model, Llama-4-Scout-17B-16E-Instruct, needs about 243 GB of VRAM at its published precision; the cheapest live fit is RTX A6000 at $2.18/hr.

Part of the Llama series · Previous generation: Llama 3.3

Pick a size

One row per official Llama 4 size: the VRAM it needs at each precision, the cheapest GPU that holds it today, and the KV cache for a 32K-token context.

ModelParametersNative VRAMFP8 VRAMINT4 VRAMLive GPU fit (native)Est. $/hrKV cache at 32K
Llama-4-Scout-17B-16E-Instruct108.6B243 GB121 GB60.7 GBRTX A6000 × 6$2.18/hrnot published for this architecture

VRAM is weights times a flat 1.2 overhead; the KV cache is a separate per-model figure at 16-bit, shown where the architecture is published. See the methodology.

Official models (1)

ModelParametersPublished asVRAM neededLive GPU fitEst. $/hr
Llama-4-Scout-17B-16E-Instruct108.6BBF16243 GBRTX A6000 × 6$2.18/hr

Quantized versions (1)

ModelParametersPublished asVRAM neededLive GPU fitEst. $/hr
Llama-4-Maverick-17B-128E-Instruct-FP8401.6BF8_E4M3449 GBRTX PRO 6000 × 5$6.88/hr

VRAM is for the precision the model is published in: the weights times a flat 1.2 overhead, with the KV cache not included. See the methodology. The fit shown is the lowest-priced single GPU type that holds the model at that precision, or the lowest-priced multi-GPU set (up to 8) when none does.

Run and fine-tune Llama 4

More from Meta

Start ComfyUI on a rented GPU.

© 2026 Aquanode. All rights reserved.

All trademarks, logos and brand names are the property of their respective owners.