olmOCR VRAM requirements
Ai2's open toolkit that turns PDFs and scanned pages into clean Markdown, including tables, equations and handwriting, with a 7 billion parameter vision language model (olmOCR-2-7B-1025, FP8 by default). Ai2 states at least 12 GB; the cheapest single GPU with that much is the 24 GB A30 at Massed Compute, $0.35 an hour.
What Ai2 says each model needs
| Model and setting | Stated GPU memory | Cheapest single GPU with that much | Per hour | |
|---|---|---|---|---|
| olmOCR-2-7B-1025Local GPU inference with vLLM on a recent NVIDIA GPU; Ai2 tested it on the RTX 4090, L40S, A100 and H100, so the card shown is Ampere or newer | at least 12 GB | A30, 24 GB, Massed Compute | $0.35 | Rent → |
The memory figures are Ai2's, not our estimate. The price is the lowest on-demand rate for one GPU with at least that much memory, from our latest check; a card that holds a model is not a promise of its speed.
Before you rent
The 12 GB is Ai2's floor for one GPU; the README also asks for 30 GB of free disk space, so check the disk that comes with the rental. If you would rather pay per token than rent a card, the README lists three hosted APIs Ai2 tested the model on: Cirrascale, DeepInfra and Parasail.
In Ai2's words
- olmOCR-2-7B-1025: “Recent NVIDIA GPU (tested on RTX 4090, L40S, A100, H100) with at least 12 GB of GPU RAM”
Source: the olmOCR README on GitHub, read 17 October 2026.
Text models are sized by a formula instead: LLM VRAM requirements for every open language model, or the LLM VRAM calculator for your own context and batch.
More models sized by their publishers' figures
- Whisper (speech recognition, OpenAI)
- Stable Diffusion 3.5 Medium (image generation, Stability AI)
- Wan2.1 (video generation, Wan-AI)
- Wan2.2 (video generation, Wan-AI)
- HunyuanVideo (video generation, Tencent)
- MinerU (OCR and document parsing, OpenDataLab)
- GTE-Qwen2 (text embeddings, Alibaba)