Update README.md

README.md

@@ -19,6 +19,27 @@ Open-source subset of the **Chunkr Reading Order benchmark dataset**, containing
 This dataset benchmarks reading order detection models on complex, real-world documents including **financial reports**, **legal contracts**, **research papers**, **medical records**, and more. Each document includes ground truth reading order sequences essential for accurate document understanding and text extraction.
 
 ---
+### Benchmark Comparison
+
+We evaluated our model against traditional heuristics, machine learning approaches, and state-of-the-art Vision-Language Models (VLMs):
+
+<div align="center">
+  <img src="analysis/overall_performance_comparison.png" alt="Model Comparison" width="100%"/>
+</div>
+
+| Model | Kendall's τ | Exact Match | Position Acc | Inference (ms) |
+|-------|------------|-------------|--------------|----------------|
+| **Chunkr Reading Order** | **97.6%** | **87.2%** | **94.8%** | **31.0** |
+| openai/gpt-5 | 83.6% | 72.0% | 80.3% | 22836.3 |
+| x-ai/grok-4-fast | 91.3% | 68.2% | 82.8% | 3886.4 |
+| Pairwise XGBoost* | 91.5% | 66.9% | 82.7% | 16.4 |
+| Pairwise LightGBM* | 91.4% | 67.0% | 82.6% | 33.8 |
+| google/gemini-2.5-pro | 80.6% | 69.9% | 77.2% | 5254.1 |
+| anthropic/claude-sonnet-4.5 | 83.6% | 53.1% | 73.0% | 1362.0 |
+| qwen/qwen3-vl-235b-a22b-instruct | 78.8% | 52.9% | 67.5% | 4732.8 |
+| XY-Cut | 82.9% | 51.6% | 68.1% | 1.0 |
+| TBLR (Baseline) | 86.8% | 47.2% | 73.2% | 1.0 |
+| z-ai/glm-4.5v | 71.2% | 43.9% | 58.4% | 8385.7 |
 
 ## Dataset Overview