Breaking

Ai2 Releases olmOCR 2 to Convert PDFs Into Markdown

June 30, 2026 at 08:15 EDT

Open Source
Foundation Models
AI Agents

The Allen Institute for Artificial Intelligence (Ai2/AllenAI) offers olmOCR, an open-source VLM toolkit that turns PDFs and scanned images into AI-friendly Markdown, and released its latest version, olmOCR 2, on October 22, 2025. The models and code are published freely, and a key selling point is local execution on your own GPU at roughly $176–178 per million pages.

October 2025 · Ai2

olmOCR 2 turns messy PDFs into clean, LLM-ready Markdown

An open-source OCR toolkit from the Allen Institute that reads tables, equations and multi-column layouts in a single VLM pass — at a fraction of the cost of closed APIs.

82.4

overall olmOCR-Bench score (±1.1)

parameter VLM, Qwen2.5-VL based

7,010

unit-test cases across 1,400 PDFs

Cost per million pages — olmOCR vs GPT-4o

olmOCR (FP8 on GPU) runs at roughly one-32nd the cost.

1×

olmOCR
<$200

32×

GPT-4o
~32× higher

Benchmark accuracy by subcategory

Tables84.9%

Multi-column layout83.7%

Old equation scans82.3%

From document to LLM-ready text — single VLM pass

PDF / PNG / JPEG

→

7B VLM
RL-tuned (GRPO)

→

Markdown + HTML tables + LaTeX

STRENGTHS

Best open-source PDF-to-Markdown converter
Outstanding on tables, equations, handwriting
Apache 2.0 — runs on local GPU or vLLM/SGLang

LIMITATIONS

No CPU support — needs NVIDIA GPU (12GB+ VRAM)
English-centric; accuracy may drop in other languages
Demo runs sequentially — batch for large volumes

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.