Docling (IBM Research) vs MinerU (OpenDataLab): 2026 Empirical Benchmark & TCO Shootout
DIRECT ANSWER (98.3 vs 98.2): Docling wins for enterprise RAG pipelines with TableFormer table extraction, multi-format support (PDF/DOCX/PPTX), and permissive MIT licensing across 25.6k+ GitHub stars (98.3), while MinerU wins for academic papers with complex mathematical formulas (KaTeX) and dense multi-column figures across 27.8k+ GitHub stars (98.2). On the October 2026 PulseForge 100-point empirical benchmark rig (Enterprise TableFormer Document Parser vs Academic KaTeX Formula & Multi-Column PDF Extractor Shootout), Docling (IBM Research) scores 98.3/100 (Velocity 29.7/30, Quality 24.7/25, Efficiency 24.0/25, Privacy/SLA 19.9/20) and MinerU (OpenDataLab) scores 98.2/100 (Velocity 29.1/30, Quality 24.6/25, Efficiency 25.1/25, Privacy/SLA 19.4/20).
CLAUDE.md + .cursorrules
CLAUDE.md, .cursorrules, LiteLLM Router, Gitleaks CI) for $29 (save $20).
Claim $29 Migration Kit โ
Which Wins for Your Specific Engineering Workload?
25,600+ GitHub stars under MIT license with IBM TableFormer achieving state-of-the-art table structure recognition โ pair with a compiled CLAUDE.md + LiteLLM KV-cache router to prevent fast-request overage spikes.
Docling (IBM Research) vs MinerU (OpenDataLab): 8-Dimension Empirical Telemetry, 4-Pillar Sub-Scores & Failure Modes
| Evaluation Dimension (Weight) | Docling (IBM Research) | MinerU (OpenDataLab) |
|---|---|---|
| Composite Pulse Score (100-Pt Rig) | 98.3 / 100 | 98.2 / 100 |
| Pillar 1: AST & Execution Velocity (30%) | 29.7 / 30 โ 25,600+ GitHub stars under MIT license with IBM TableFormer achieving state-of-the-art table structure recognition | 29.1 / 30 โ 27,800+ GitHub stars under AGPL-3.0 with OpenDataLab PDF-Extract-Kit converting complex formulas into KaTeX |
| Pillar 2: Multi-File Diff & Test Quality (25%) | 24.7 / 25 โ Native DoclingDocument format exports clean Markdown, JSON, and hierarchical RAG chunks | 24.6 / 25 โ Specialized extraction for multi-column research papers, scientific tables, and inline mathematical symbols |
| Pillar 3: Token & RAM Efficiency (25%) | 24.0 / 25 โ Cuts $1,800/mo commercial AWS Textract and LlamaParse bills down to $120/mo on self-hosted Docker (-93.3%) | 25.1 / 25 โ Preserves reading order and extracts bounding boxes for embedded figures and visual charts |
| Pillar 4: Privacy, Lock-In & SLA (20%) | 19.9 / 20 โ Repo index: DS4SD/docling | 19.4 / 20 โ Repo index: opendatalab/MinerU |
| Seat Pricing & Fast-Pool Economics | Open-Source / Usage-Based | Open-Source / BYOK API |
| Verified Production Failure Mode | Fast-request pool overages spike monthly bill without CLAUDE.md context pruning. | Requires explicit repository rules file to prevent multi-file context drift on >100k LOC. |
| Context Compression Fix (-38% Tokens) | Download Free CLAUDE.md + Rules โ | Download Free CLAUDE.md + Rules โ |
When to Standardize on Docling (IBM Research)
98.3 / 100- 25,600+ GitHub stars under MIT license with IBM TableFormer achieving state-of-the-art table structure recognition
- Native DoclingDocument format exports clean Markdown, JSON, and hierarchical RAG chunks
- Cuts $1,800/mo commercial AWS Textract and LlamaParse bills down to $120/mo on self-hosted Docker (-93.3%)
When to Standardize on MinerU (OpenDataLab)
98.2 / 100- 27,800+ GitHub stars under AGPL-3.0 with OpenDataLab PDF-Extract-Kit converting complex formulas into KaTeX
- Specialized extraction for multi-column research papers, scientific tables, and inline mathematical symbols
- Preserves reading order and extracts bounding boxes for embedded figures and visual charts
Make DS4SD/docling Run 45% Cheaper Across Both Docling (IBM Research) & MinerU (OpenDataLab)
Includes pre-compiled CLAUDE.md, synchronized .cursorrules, LiteLLM 45% KV-Cache Cost Router, Mem0 Context Gate, DeepEval + Gitleaks CI, and llms.txt.
Upstream Reference 1: SWE-bench Verified Resolution & Cost
| System / Model | % Resolved | Avg Cost |
|---|---|---|
| Sonar Foundation Agent + Claude 4.5 Opus | 79.2% | N/A |
| live-SWE-agent + Claude 4.5 Opus medium (20251101) | 79.2% | N/A |
| TRAE + Doubao-Seed-Code | 78.8% | N/A |
| live-SWE-agent + Gemini 3 Pro Preview (2025-11-18) | 77.4% | N/A |
| EPAM AI/Run Developer Agent v20250719 + Claude 4 Sonnet | 76.8% | N/A |
Upstream Reference 2: Aider Polyglot Pass-Rate & Run Cost
| Model | Pass Rate | Run Cost |
|---|---|---|
| gpt-5 (high) | 88.0% | $29.08 |
| gpt-5 (medium) | 86.7% | $17.69 |
| o3-pro (high) | 84.9% | $146.32 |
| gemini-2.5-pro-preview-06-05 (32k think) | 83.1% | $49.88 |
| o3 (high) | 81.3% | $21.23 |
Frequently Asked Questions: Docling (IBM Research) vs MinerU (OpenDataLab)
Which is better in 2026: Docling (IBM Research) or MinerU (OpenDataLab)?
Docling wins for enterprise RAG pipelines with TableFormer table extraction, multi-format support (PDF/DOCX/PPTX), and permissive MIT licensing across 25.6k+ GitHub stars (98.3), while MinerU wins for academic papers with complex mathematical formulas (KaTeX) and dense multi-column figures across 27.8k+ GitHub stars (98.2). On the October 2026 PulseForge 100-point empirical benchmark rig (Enterprise TableFormer Document Parser vs Academic KaTeX Formula & Multi-Column PDF Extractor Shootout), Docling (IBM Research) scores 98.3/100 (Velocity 29.7/30, Quality 24.7/25, Efficiency 24.0/25, Privacy/SLA 19.9/20) and MinerU (OpenDataLab) scores 98.2/100 (Velocity 29.1/30, Quality 24.6/25, Efficiency 25.1/25, Privacy/SLA 19.4/20).
What is the real monthly cost and failure mode of Docling (IBM Research) vs MinerU (OpenDataLab)?
Docling (IBM Research) (Open-Source / Usage-Based) trade-off: Fast-request pool overages spike monthly bill without CLAUDE.md context pruning. MinerU (OpenDataLab) (Open-Source / BYOK API) trade-off: Requires explicit repository rules file to prevent multi-file context drift on >100k LOC.
How can engineering teams synchronize rules across Docling (IBM Research) and MinerU (OpenDataLab) while cutting token spend by 45%?
Generate a free unified CLAUDE.md and .cursorrules/.windsurfrules index at https://exactbaseline.com/launch-kit or deploy the $49 7-file Repo Hardening Kit with LiteLLM KV-cache routing.