Granite-Docling
Granite-Docling is an open-weights language model from IBM, released September 17, 2025. Its makers disclosed about 0.3B parameters.
Today, IBM is releasing Granite-Docling-258M, an ultra-compact and cutting-edge open-source vision-language model (VLM) for converting documents to machine-readable formats while fully preserving their layout, tables, equations, lists and more. It’s now available on Hugging Face through a standard Apache 2.0 license.
Check it against your own PC →
Can you run Granite-Docling locally?
Even with 32 GB of VRAM and 32 GB of RAM it is too large to run at a usable speed on a single consumer GPU. Estimates assume 32 GB of DDR5 RAM, a 16K context and llama.cpp-style GGUF quantization. Enter your exact hardware in the live app.
| GPU memory | Verdict | Best quant | VRAM | RAM | tok/s |
|---|---|---|---|---|---|
| 6 GB VRAM | Size unknown | – | – | – | – |
| 8 GB VRAM | Size unknown | – | – | – | – |
| 12 GB VRAM | Size unknown | – | – | – | – |
| 16 GB VRAM | Size unknown | – | – | – | – |
| 24 GB VRAM | Size unknown | – | – | – | – |
| 32 GB VRAM | Size unknown | – | – | – | – |
Parameter count unknown. The weights may not be on Hugging Face yet.
Benchmark results
Epoch AI has not published results for Granite-Docling yet. New models are usually evaluated within days to weeks.
Links
More from IBM
granite switch 4.1 8b · granite switch 4.1 30b · Granite 4.1 30B · Granite 4.1 8B · Granite-4.0-H-Tiny · Granite-4.0-H-Micro · Granite-4.0-H-Small