FrontierThe story, in brief
Datalab Marker v2 vs MinerU, Docling, and Liteparse: Benchmark Breakdown
76.0 on olmOCR-bench. Datalab's Marker v2 just hit 5× faster than MinerU while beating Docling on accuracy.
The KeyNews take
Why it matters
Document parsing pipelines are becoming a critical infrastructure layer for enterprise AI workflows. Marker v2's performance gains matter because they directly impact the cost and speed of RAG systems and document-heavy LLM applications at scale.
The key facts
6 to knowMarker v2 scores 76.0 on olmOCR-bench
Sustains 2.9 pages per second on single B200
5× faster than MinerU's pipeline backend
Beats Docling on both accuracy and speed
Three-mode pipeline architecture rewrite
Compared against: MinerU, Docling, LiteParse
Go to the source
MarkTechPostmarktechpost.com
Publisher excerpt: Datalab rewrote Marker as a three-mode pipeline. Version 2 hits 76.0 on olmOCR-bench and sustains 2.9 pages per second on one B200 — over 5× MinerU's pipeline backend, while beating Docling on both accuracy and speed. Here's how it compares against MinerU, Docling and LiteParse, and which one fits…