Datalab’s Marker 2 vs MinerU, Docling and LiteParse: 76.0 on olmOCR-bench at 5× MinerU’s Throughput

Kwon Crash

Published Jul 25, 2026, 6:00 AM UTC

Source: AISource
- Datalab’s Marker 2 just slapped MinerU and Docling with a 5x throughput reality check. While the rest of the industry is busy minting JPEGs of nothing, this open-source pipeline is churning out markdown at 2.9 pages per second on a B200. It’s faster, more accurate, and doesn’t require a VC’s blessing to run. If you’re still using legacy OCR that moves slower than a Core Dynamics audit, you’re basically leaving money on the table. This is the kind of raw digital infrastructure efficiency we need—no hype, just speed. Stop treating your data like unsealed cargo; upgrade your stack or get left in the dust. Where's my cut?