Granite 4 models

22 Granite 4 models from IBM on Hugging Face, from 352M to 32.2B parameters, published by IBM in BF16, with contexts from 4K to 128K tokens. The smallest official model, granite-4.0-350m, needs about 0.8 GB of VRAM at its published precision; the cheapest live fit is RTX 3070 at $0.088/hr.

Part of the Granite series ยท Previous generation: Granite 3

Pick a size

One row per official Granite 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
granite-4.0-350m352M0.8 GB0.4 GB0.2 GBRTX 3070$0.088/hr0.88 GB
granite-4.0-1b-base1.6B3.6 GB1.8 GB0.9 GBRTX 3070$0.088/hr2.5 GB
granite-4.0-1b-speech2.3B5.2 GB2.6 GB1.3 GBRTX 3070$0.088/hr2.5 GB
granite-4.0-h-micro3.2B7.1 GB3.6 GB1.8 GBRTX 3070$0.088/hr0.25 GB
granite-4.1-3b3.4B7.6 GB3.8 GB1.9 GBRTX 3070$0.088/hr2.5 GB
granite-4.2-3b3.7B8.2 GB4.1 GB2.0 GBRTX 3060$0.110/hr2.5 GB
granite-vision-4.1-4b4.0B8.9 GB4.5 GB2.2 GBRTX 3060$0.110/hr2.5 GB
granite-4.0-h-tiny6.9B15.5 GB7.8 GB3.9 GBRTX A4000$0.167/hr0.25 GB
granite-guardian-4.1-8b8.4B18.7 GB9.4 GB4.7 GBRTX A5000$0.176/hr5.0 GB
granite-4.1-8b8.8B19.7 GB9.8 GB4.9 GBRTX A5000$0.176/hr5.0 GB
granite-4.1-30b28.9B64.5 GB32.3 GB16.1 GBA100$1.31/hr8.0 GB
granite-4.2-30b29.3B65.4 GB32.7 GB16.4 GBA100$1.31/hr8.0 GB
granite-4.0-h-small32.2B72.0 GB36.0 GB18.0 GBA100$1.31/hr0.50 GB

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 (18)

ModelParametersPublished asVRAM neededLive GPU fitEst. $/hr
granite-4.0-350m352MBF160.8 GBRTX 3070$0.088/hr
granite-4.0-1b1.6BBF163.6 GBRTX 3070$0.088/hr
granite-4.0-1b-base1.6BBF163.6 GBRTX 3070$0.088/hr
granite-4.0-1b-speech2.3BBF165.2 GBRTX 3070$0.088/hr
granite-4.0-h-micro3.2BBF167.1 GBRTX 3070$0.088/hr
granite-4.0-micro3.4BBF167.6 GBRTX 3070$0.088/hr
granite-4.1-3b3.4BBF167.6 GBRTX 3070$0.088/hr
granite-4.1-3b-llmon-preview3.7BBF168.2 GBRTX 3060$0.110/hr
granite-4.2-3b3.7BBF168.2 GBRTX 3060$0.110/hr
granite-vision-4.1-4b4.0BBF168.9 GBRTX 3060$0.110/hr
granite-4.0-h-tiny6.9BBF1615.5 GBRTX A4000$0.167/hr
granite-guardian-4.1-8b8.4BBF1618.7 GBRTX A5000$0.176/hr
granite-4.1-8b8.8BBF1619.7 GBRTX A5000$0.176/hr
granite-4.1-8b-llmon-preview8.8BBF1619.7 GBRTX A5000$0.176/hr
granite-4.2-8b8.8BBF1619.7 GBRTX A5000$0.176/hr
granite-4.1-30b28.9BBF1664.5 GBA100$1.31/hr
granite-4.2-30b29.3BBF1665.4 GBA100$1.31/hr
granite-4.0-h-small32.2BBF1672.0 GBA100$1.31/hr

Quantized versions (1)

ModelParametersPublished asVRAM neededLive GPU fitEst. $/hr
granite-4.1-8b-fp88.8BF8_E4M39.8 GBRTX 4070$0.154/hr

Fine-tunes and community models (3)

ModelParametersPublished asVRAM neededLive GPU fitEst. $/hr
antares-1b1.8BBF164.1 GBRTX 3070$0.088/hr
TwIL-LM3-Pro3.7BBF168.2 GBRTX 3060$0.110/hr
TwIL-8B8.8BBF1619.7 GBRTX A5000$0.176/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 Granite 4

More from IBM

Start ComfyUI on a rented GPU.

ยฉ 2026 Aquanode. All rights reserved.

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