Datalab Marker v2 vs MinerU, Docling, and Liteparse: Benchmark Breakdown

Document processing technology has reached a critical inflection point with Datalab's release of Marker v2, a substantially rewritten pipeline that demonstrates significant performance improvements over competing solutions. The latest version combines enhanced accuracy metrics with dramatic speed improvements, establishing new baseline expectations for document conversion and analysis at scale.

Datalab's Marker v2 operates on a three-mode pipeline architecture and achieves a 76.0 score on the olmOCR-bench benchmark while maintaining a processing speed of 2.9 pages per second on a single B200 GPU. This throughput represents a more than fivefold improvement over MinerU's pipeline backend performance. The tool simultaneously surpasses Docling in both accuracy measurements and processing velocity, marking a substantial advancement in the document conversion landscape. These benchmarks were conducted under controlled conditions designed to reflect real-world document processing scenarios.

Key performance differentiators include:

• 76.0 olmOCR-bench accuracy score, outperforming comparable solutions
• 2.9 pages per second processing speed on B200 hardware (>5× faster than MinerU backend)
• Superior combined accuracy and speed metrics versus Docling
• Three-mode pipeline design enabling flexible deployment options
• Sustained performance improvements addressing previous bottlenecks in document parsing

The performance leap demonstrated by Marker v2 carries significant implications for organizations managing large-scale document processing workloads. Enterprises previously constrained by processing speed limitations now have viable options for real-time or near-real-time document conversion pipelines. The substantial performance gap compared to established competitors may accelerate migration efforts and influence technology selection for new projects.

Organizations must weigh Marker v2's performance advantages against integration requirements, existing infrastructure compatibility, and specific use-case demands. While raw speed and accuracy represent critical metrics, implementation complexity and maintenance overhead require careful evaluation alongside benchmark results.

Marker v2's achievement represents a meaningful evolution in document processing technology. As organizations increasingly rely on automated document conversion for AI pipelines, data extraction, and content management systems, performance benchmarks become central to infrastructure decisions. Datalab's rewritten pipeline raises the performance bar industry-wide, compelling competitors to match or exceed these metrics while pushing the technological boundaries of what document processing pipelines can achieve at scale.