Publish initial MDPBench leaderboard

.gitignore

@@ -0,0 +1,3 @@
+__pycache__/
+*.pyc
+.DS_Store

README.md

@@ -1,10 +1,39 @@
 ---
 title: MDPBench Leaderboard
-emoji: 📊
-colorFrom: red
-colorTo: green
+emoji: 📄
+colorFrom: blue
+colorTo: yellow
 sdk: static
 pinned: false
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# MDPBench Leaderboard
+
+Static, official leaderboard for **MDPBench: A Benchmark for Multilingual
+Document Parsing in Real-World Scenarios**.
+
+The deployed page needs no server or API key. Update `leaderboard.json` to
+publish a new verified result. The initial entries are the results in the
+repository's main-results table; `source_name` keeps the original table label
+where the displayed official release/API identifier is more specific. The
+initial board contains all 31 models from that table.
+
+Before publishing an updated result table, synchronize the data and verify the
+result:
+
+```bash
+python tools/sync_leaderboard.py --write
+python tools/sync_leaderboard.py --check
+```
+
+## Deployment
+
+Create a Hugging Face Space with the **Static** SDK and upload the contents of
+this directory. Its `README.md` metadata configures the Space automatically.
+
+## Result policy
+
+Public scores can be reproduced using the released evaluation set. Private
+scores are evaluated by the MDPBench maintainers; a model is ranked by private
+score only after that verification. See the benchmark README for the private-set
+submission procedure.

app.js

@@ -0,0 +1,56 @@
+const state = { sort: "overall", query: "", type: "all" };
+const languages = ["DE", "EN", "ES", "FR", "ID", "IT", "NL", "PT", "VI", "AR", "HI", "JP", "KO", "RU", "TH", "ZH", "ZH-T"];
+
+function value(score) {
+  return Number.isFinite(score) ? score.toFixed(1) : "—";
+}
+
+function metadata(model) {
+  return MODEL_METADATA[model.name] || { label: model.name, url: null };
+}
+
+function makeDetail(model) {
+  const rows = languages.map((language) => `<span><b>${language}</b>${value(model.languages[language])}</span>`).join("");
+  const source = model.source_name ? `<p class="source-note">Benchmark table label: <code>${model.source_name}</code></p>` : "";
+  return `<tr class="detail-row"><td colspan="10"><div class="details"><div><p class="detail-title">Language-level public scores</p><div class="language-grid">${rows}</div></div>${source}</div></td></tr>`;
+}
+
+function render(data) {
+  const models = data.models
+    .filter((m) => state.type === "all" || m.type === state.type)
+    .filter((m) => `${m.name} ${m.source_name || ""}`.toLowerCase().includes(state.query))
+    .sort((a, b) => b[state.sort] - a[state.sort]);
+  const body = document.querySelector("#leaderboard-body");
+  body.innerHTML = models.map((model, index) => `
+    <tr class="model-row">
+      <td class="rank">${index + 1}</td><td><span class="model-name">${metadata(model).label}</span>${metadata(model).url ? `<a class="model-link" href="${metadata(model).url}" target="_blank" rel="noreferrer" aria-label="Open official model page for ${metadata(model).label}">↗</a>` : ""}</td>
+      <td><span class="type type-${model.type.toLowerCase().replaceAll(" ", "-")}">${model.type}</span></td>
+      <td class="score featured">${value(model.overall)}</td><td class="score">${value(model.digital)}</td><td class="score">${value(model.photo)}</td>
+      <td class="score">${value(model.latin)}</td><td class="score">${value(model.non_latin)}</td><td class="score private">${value(model.private)}</td>
+      <td><button class="expand" aria-label="Show language scores">+</button></td>
+    </tr>${makeDetail(model)}`).join("");
+  body.querySelectorAll(".model-row").forEach((row) => {
+    const toggle = () => {
+      const detail = row.nextElementSibling;
+      const open = detail.classList.toggle("show");
+      row.querySelector(".expand").textContent = open ? "−" : "+";
+      row.querySelector(".model-name").setAttribute("aria-expanded", String(open));
+    };
+    row.querySelector(".expand").addEventListener("click", toggle);
+  });
+}
+
+fetch("leaderboard.json")
+  .then((response) => response.json())
+  .then((data) => {
+    document.querySelector("#updated").textContent = `Last updated · ${data.benchmark.updated}`;
+    render(data);
+    document.querySelectorAll("[data-sort]").forEach((button) => button.addEventListener("click", () => {
+      state.sort = button.dataset.sort;
+      document.querySelectorAll("[data-sort]").forEach((item) => item.classList.toggle("active", item === button));
+      render(data);
+    }));
+    document.querySelector("#search").addEventListener("input", (event) => { state.query = event.target.value.toLowerCase().trim(); render(data); });
+    document.querySelector("#type-filter").addEventListener("change", (event) => { state.type = event.target.value; render(data); });
+  })
+  .catch(() => { document.querySelector("#leaderboard-body").innerHTML = '<tr><td colspan="10" class="error">Could not load leaderboard data.</td></tr>'; });

index.html

@@ -1,19 +1,69 @@
 <!doctype html>
-<html>
-	<head>
-		<meta charset="utf-8" />
-		<meta name="viewport" content="width=device-width" />
-		<title>My static Space</title>
-		<link rel="stylesheet" href="style.css" />
-	</head>
-	<body>
-		<div class="card">
-			<h1>Welcome to your static Space!</h1>
-			<p>You can modify this app directly by editing <i>index.html</i> in the Files and versions tab.</p>
-			<p>
-				Also don't forget to check the
-				<a href="https://huggingface.co/docs/hub/spaces" target="_blank">Spaces documentation</a>.
-			</p>
-		</div>
-	</body>
+<html lang="en">
+  <head>
+    <meta charset="utf-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1" />
+    <title>MDPBench Leaderboard</title>
+    <meta name="description" content="Official leaderboard for MDPBench, a benchmark for multilingual document parsing in real-world scenarios." />
+    <link rel="preconnect" href="https://fonts.googleapis.com" />
+    <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin />
+    <link href="https://fonts.googleapis.com/css2?family=DM+Mono&family=Manrope:wght@400;500;600;700;800&display=swap" rel="stylesheet" />
+    <link rel="stylesheet" href="style.css" />
+  </head>
+  <body>
+    <main class="shell">
+      <header class="hero">
+        <a class="brand" href="https://github.com/Yuliang-Liu/MultimodalOCR" target="_blank" rel="noreferrer">MDP<span>Bench</span></a>
+        <p class="eyebrow">Official leaderboard · Multilingual document parsing</p>
+        <h1>How well do models parse<br /><em>documents in the real world?</em></h1>
+        <p class="intro">A unified evaluation across 17 languages, native digital pages, and photographed documents. Scores are shown as percentages; higher is better.</p>
+        <div class="links">
+          <a href="https://arxiv.org/abs/2603.28130" target="_blank" rel="noreferrer">Paper ↗</a>
+          <a href="https://huggingface.co/datasets/Delores-Lin/MDPBench" target="_blank" rel="noreferrer">Dataset ↗</a>
+          <a href="https://github.com/Yuliang-Liu/MultimodalOCR" target="_blank" rel="noreferrer">GitHub ↗</a>
+        </div>
+      </header>
+
+      <section class="at-a-glance" aria-label="Benchmark facts">
+        <div><strong>3,400</strong><span>document images</span></div>
+        <div><strong>17</strong><span>languages</span></div>
+        <div><strong>3</strong><span>parsing tasks</span></div>
+        <div><strong>2</strong><span>evaluation splits</span></div>
+      </section>
+
+      <section class="board" aria-labelledby="board-title">
+        <div class="board-topline">
+          <div>
+            <p class="eyebrow">Verified results</p>
+            <h2 id="board-title">Leaderboard</h2>
+          </div>
+          <p id="updated" class="updated"></p>
+        </div>
+
+        <div class="controls">
+          <div class="rank-tabs" role="group" aria-label="Ranking score">
+            <button class="active" data-sort="overall">Public Overall</button>
+            <button data-sort="private">Private Overall</button>
+          </div>
+          <label class="search"><span class="sr-only">Search models</span><input id="search" type="search" placeholder="Search model" /></label>
+          <label class="select-label"><span>Model family</span><select id="type-filter"><option value="all">All types</option><option>General VLM</option><option>Specialized VLM</option><option>Pipeline Tool</option></select></label>
+        </div>
+
+        <div class="table-wrap">
+          <table>
+            <thead><tr><th>Rank</th><th>Model</th><th>Type</th><th>Overall</th><th>Digital</th><th>Photo</th><th>Latin</th><th>Non-Latin</th><th>Private</th><th aria-label="Details"></th></tr></thead>
+            <tbody id="leaderboard-body"></tbody>
+          </table>
+        </div>
+      </section>
+
+      <section class="methodology">
+        <div><p class="eyebrow">How to read this board</p><h2>One number, with the context still visible.</h2></div>
+        <div class="method-copy"><p>The public score is calculated over released samples. Private scores use the held-out split and are the basis for verified comparison. Digital and Photo expose the real-world robustness gap; Latin and Non-Latin avoid hiding language imbalance behind an average.</p><p>To add a model, run the released evaluation pipeline, then provide inference code, weights or API details, and predictions for official private-set evaluation.</p></div>
+      </section>
+    </main>
+    <footer>MDPBench · Results are maintained by the benchmark authors.</footer>
+    <script src="model_metadata.js"></script>
+    <script src="app.js"></script>
+  </body>
 </html>

leaderboard.json

@@ -0,0 +1,918 @@
+{
+  "benchmark": {
+    "name": "MDPBench",
+    "version": "Initial official results",
+    "updated": "2026-07-20",
+    "metric": "Page-level mean of available text, formula, and table task scores (×100)",
+    "private_policy": "Private scores are run and verified by the MDPBench maintainers."
+  },
+  "models": [
+    {
+      "type": "General VLM",
+      "name": "gemini-3-pro-preview",
+      "overall": 86.4,
+      "digital": 90.4,
+      "photo": 85.1,
+      "latin": 88.4,
+      "non_latin": 84.1,
+      "private": 89.8,
+      "languages": {
+        "DE": 91.2,
+        "EN": 90.6,
+        "ES": 83.4,
+        "FR": 82.7,
+        "ID": 91.5,
+        "IT": 91.6,
+        "NL": 87.7,
+        "PT": 91.4,
+        "VI": 85.9,
+        "AR": 89.4,
+        "HI": 90.4,
+        "JP": 74.8,
+        "KO": 85.5,
+        "RU": 84.9,
+        "TH": 80.6,
+        "ZH": 85.1,
+        "ZH-T": 82.1
+      },
+      "source_name": "Gemini-3-pro-preview"
+    },
+    {
+      "type": "General VLM",
+      "name": "kimi-k2.5",
+      "overall": 77.5,
+      "digital": 85.0,
+      "photo": 75.0,
+      "latin": 81.6,
+      "non_latin": 72.9,
+      "private": 81.2,
+      "languages": {
+        "DE": 85.9,
+        "EN": 86.2,
+        "ES": 72.7,
+        "FR": 71.0,
+        "ID": 80.6,
+        "IT": 86.6,
+        "NL": 77.4,
+        "PT": 87.6,
+        "VI": 86.2,
+        "AR": 75.8,
+        "HI": 74.5,
+        "JP": 72.5,
+        "KO": 70.9,
+        "RU": 61.8,
+        "TH": 67.0,
+        "ZH": 81.7,
+        "ZH-T": 78.6
+      },
+      "source_name": "kimi-K2.5"
+    },
+    {
+      "type": "General VLM",
+      "name": "doubao-seed-2-0-pro-260215",
+      "overall": 74.2,
+      "digital": 78.9,
+      "photo": 72.8,
+      "latin": 75.7,
+      "non_latin": 72.5,
+      "private": 79.5,
+      "languages": {
+        "DE": 82.8,
+        "EN": 74.4,
+        "ES": 69.0,
+        "FR": 70.0,
+        "ID": 73.3,
+        "IT": 82.0,
+        "NL": 69.9,
+        "PT": 83.4,
+        "VI": 76.5,
+        "AR": 81.3,
+        "HI": 75.7,
+        "JP": 65.8,
+        "KO": 74.7,
+        "RU": 63.3,
+        "TH": 71.9,
+        "ZH": 71.9,
+        "ZH-T": 75.2
+      },
+      "source_name": "Doubao-2.0-pro"
+    },
+    {
+      "type": "General VLM",
+      "name": "claude-sonnet-4-6",
+      "overall": 73.1,
+      "digital": 85.0,
+      "photo": 69.3,
+      "latin": 79.2,
+      "non_latin": 66.2,
+      "private": 77.6,
+      "languages": {
+        "DE": 79.8,
+        "EN": 80.6,
+        "ES": 72.8,
+        "FR": 66.5,
+        "ID": 82.3,
+        "IT": 83.3,
+        "NL": 76.7,
+        "PT": 88.0,
+        "VI": 83.1,
+        "AR": 67.8,
+        "HI": 71.7,
+        "JP": 63.4,
+        "KO": 64.3,
+        "RU": 70.8,
+        "TH": 65.2,
+        "ZH": 61.3,
+        "ZH-T": 65.1
+      },
+      "source_name": "Claude-Sonnet-4.6"
+    },
+    {
+      "type": "General VLM",
+      "name": "gpt-5.2-2025-12-11",
+      "overall": 68.6,
+      "digital": 85.6,
+      "photo": 63.0,
+      "latin": 75.2,
+      "non_latin": 61.1,
+      "private": 74.0,
+      "languages": {
+        "DE": 70.8,
+        "EN": 79.4,
+        "ES": 71.4,
+        "FR": 60.0,
+        "ID": 77.7,
+        "IT": 78.5,
+        "NL": 71.6,
+        "PT": 85.0,
+        "VI": 82.1,
+        "AR": 64.9,
+        "HI": 63.4,
+        "JP": 55.8,
+        "KO": 65.4,
+        "RU": 60.7,
+        "TH": 63.8,
+        "ZH": 56.3,
+        "ZH-T": 58.7
+      },
+      "source_name": "ChatGPT-5.2-2025-12-11"
+    },
+    {
+      "type": "General VLM",
+      "name": "Qwen3-VL-8B-Instruct",
+      "overall": 68.3,
+      "digital": 78.4,
+      "photo": 65.0,
+      "latin": 73.6,
+      "non_latin": 62.5,
+      "private": 70.8,
+      "languages": {
+        "DE": 73.7,
+        "EN": 71.4,
+        "ES": 69.3,
+        "FR": 66.2,
+        "ID": 68.5,
+        "IT": 79.1,
+        "NL": 78.3,
+        "PT": 82.2,
+        "VI": 73.4,
+        "AR": 63.1,
+        "HI": 58.4,
+        "JP": 59.9,
+        "KO": 61.9,
+        "RU": 57.9,
+        "TH": 62.0,
+        "ZH": 62.6,
+        "ZH-T": 73.8
+      },
+      "source_name": "Qwen3-VL-Instruct-8b"
+    },
+    {
+      "type": "General VLM",
+      "name": "Qwen3.5-Instruct-9B",
+      "overall": 65.7,
+      "digital": 74.8,
+      "photo": 62.7,
+      "latin": 72.5,
+      "non_latin": 58.2,
+      "private": 68.9,
+      "languages": {
+        "DE": 72.8,
+        "EN": 72.0,
+        "ES": 72.0,
+        "FR": 64.4,
+        "ID": 66.2,
+        "IT": 77.6,
+        "NL": 74.5,
+        "PT": 79.1,
+        "VI": 74.0,
+        "AR": 53.4,
+        "HI": 56.2,
+        "JP": 55.7,
+        "KO": 60.3,
+        "RU": 54.7,
+        "TH": 56.7,
+        "ZH": 60.8,
+        "ZH-T": 67.5
+      }
+    },
+    {
+      "type": "General VLM",
+      "name": "InternVL3_5-8B",
+      "overall": 42.7,
+      "digital": 59.7,
+      "photo": 37.0,
+      "latin": 53.4,
+      "non_latin": 30.6,
+      "private": 45.3,
+      "languages": {
+        "DE": 39.8,
+        "EN": 64.2,
+        "ES": 47.5,
+        "FR": 42.7,
+        "ID": 53.8,
+        "IT": 60.6,
+        "NL": 52.2,
+        "PT": 63.2,
+        "VI": 57.0,
+        "AR": 8.2,
+        "HI": 9.0,
+        "JP": 45.6,
+        "KO": 30.3,
+        "RU": 26.1,
+        "TH": 10.8,
+        "ZH": 55.3,
+        "ZH-T": 59.3
+      },
+      "source_name": "InternVL-3.5-8B"
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "MonkeyOCRv2-B-Parsing",
+      "overall": 83.3,
+      "digital": 88.1,
+      "photo": 81.7,
+      "latin": 84.2,
+      "non_latin": 82.1,
+      "private": 84.7,
+      "languages": {
+        "DE": 87.7,
+        "EN": 84.5,
+        "ES": 75.2,
+        "FR": 78.4,
+        "ID": 86.5,
+        "IT": 88.6,
+        "NL": 86.1,
+        "PT": 87.9,
+        "VI": 83.2,
+        "AR": 90.7,
+        "HI": 87.2,
+        "JP": 71.9,
+        "KO": 87.6,
+        "RU": 80.1,
+        "TH": 80.8,
+        "ZH": 83.6,
+        "ZH-T": 75.3
+      }
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "MonkeyOCRv2-S-Parsing",
+      "overall": 82.5,
+      "digital": 87.9,
+      "photo": 80.7,
+      "latin": 83.2,
+      "non_latin": 81.7,
+      "private": 84.0,
+      "languages": {
+        "DE": 87.3,
+        "EN": 83.6,
+        "ES": 76.8,
+        "FR": 73.6,
+        "ID": 85.4,
+        "IT": 87.2,
+        "NL": 85.5,
+        "PT": 87.4,
+        "VI": 81.9,
+        "AR": 91.2,
+        "HI": 87.1,
+        "JP": 69.9,
+        "KO": 88.7,
+        "RU": 78.0,
+        "TH": 79.8,
+        "ZH": 84.4,
+        "ZH-T": 74.7
+      }
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "dots.mocr",
+      "overall": 80.5,
+      "digital": 90.5,
+      "photo": 77.2,
+      "latin": 81.7,
+      "non_latin": 79.2,
+      "private": 82.8,
+      "languages": {
+        "DE": 82.6,
+        "EN": 87.4,
+        "ES": 71.3,
+        "FR": 70.1,
+        "ID": 84.5,
+        "IT": 89.3,
+        "NL": 83.2,
+        "PT": 86.8,
+        "VI": 79.9,
+        "AR": 83.3,
+        "HI": 83.6,
+        "JP": 75.0,
+        "KO": 78.7,
+        "RU": 71.2,
+        "TH": 77.9,
+        "ZH": 84.6,
+        "ZH-T": 79.6
+      }
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "chandra-ocr-2",
+      "overall": 79.7,
+      "digital": 87.8,
+      "photo": 77.1,
+      "latin": 82.7,
+      "non_latin": 76.4,
+      "private": 82.4,
+      "languages": {
+        "DE": 86.6,
+        "EN": 86.5,
+        "ES": 69.7,
+        "FR": 70.3,
+        "ID": 84.6,
+        "IT": 87.4,
+        "NL": 82.7,
+        "PT": 90.7,
+        "VI": 85.6,
+        "AR": 78.2,
+        "HI": 81.1,
+        "JP": 68.8,
+        "KO": 80.3,
+        "RU": 74.0,
+        "TH": 78.5,
+        "ZH": 73.8,
+        "ZH-T": 76.3
+      }
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "PaddleOCR-VL-1.5",
+      "overall": 78.3,
+      "digital": 87.4,
+      "photo": 75.2,
+      "latin": 81.2,
+      "non_latin": 74.9,
+      "private": 80.7,
+      "languages": {
+        "DE": 84.8,
+        "EN": 83.0,
+        "ES": 75.7,
+        "FR": 78.1,
+        "ID": 83.9,
+        "IT": 85.2,
+        "NL": 80.6,
+        "PT": 80.2,
+        "VI": 78.9,
+        "AR": 71.3,
+        "HI": 67.7,
+        "JP": 69.5,
+        "KO": 86.0,
+        "RU": 76.0,
+        "TH": 68.4,
+        "ZH": 84.8,
+        "ZH-T": 75.7
+      }
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "HunyuanOCR-1.5",
+      "overall": 76.8,
+      "digital": 86.2,
+      "photo": 73.6,
+      "latin": 79.7,
+      "non_latin": 73.5,
+      "private": 81.2,
+      "languages": {
+        "DE": 79.6,
+        "EN": 80.4,
+        "ES": 74.2,
+        "FR": 70.0,
+        "ID": 81.5,
+        "IT": 84.5,
+        "NL": 78.4,
+        "PT": 86.4,
+        "VI": 82.4,
+        "AR": 71.8,
+        "HI": 71.6,
+        "JP": 65.5,
+        "KO": 75.7,
+        "RU": 67.4,
+        "TH": 77.7,
+        "ZH": 80.8,
+        "ZH-T": 77.2
+      }
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "dots.ocr",
+      "overall": 76.5,
+      "digital": 88.8,
+      "photo": 72.3,
+      "latin": 79.1,
+      "non_latin": 73.5,
+      "private": 79.7,
+      "languages": {
+        "DE": 79.7,
+        "EN": 81.2,
+        "ES": 69.2,
+        "FR": 67.1,
+        "ID": 82.5,
+        "IT": 87.8,
+        "NL": 78.8,
+        "PT": 86.9,
+        "VI": 79.1,
+        "AR": 75.9,
+        "HI": 77.3,
+        "JP": 70.6,
+        "KO": 68.5,
+        "RU": 66.8,
+        "TH": 73.3,
+        "ZH": 79.1,
+        "ZH-T": 76.2
+      }
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "PaddleOCR-VL-1.6",
+      "overall": 75.0,
+      "digital": 82.8,
+      "photo": 72.6,
+      "latin": 78.0,
+      "non_latin": 71.6,
+      "private": 77.0,
+      "languages": {
+        "DE": 84.1,
+        "EN": 79.7,
+        "ES": 69.2,
+        "FR": 74.8,
+        "ID": 81.6,
+        "IT": 82.0,
+        "NL": 74.7,
+        "PT": 76.4,
+        "VI": 79.3,
+        "AR": 69.4,
+        "HI": 65.6,
+        "JP": 68.7,
+        "KO": 82.5,
+        "RU": 70.7,
+        "TH": 62.3,
+        "ZH": 78.0,
+        "ZH-T": 75.7
+      }
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "MinerU-2.5-Pro-1.2B",
+      "overall": 71.0,
+      "digital": 86.2,
+      "photo": 66.1,
+      "latin": 74.6,
+      "non_latin": 67.0,
+      "private": 74.3,
+      "languages": {
+        "DE": 78.3,
+        "EN": 79.5,
+        "ES": 63.4,
+        "FR": 67.4,
+        "ID": 78.0,
+        "IT": 79.7,
+        "NL": 72.1,
+        "PT": 78.6,
+        "VI": 74.2,
+        "AR": 56.6,
+        "HI": 72.2,
+        "JP": 59.1,
+        "KO": 77.6,
+        "RU": 62.6,
+        "TH": 61.8,
+        "ZH": 76.5,
+        "ZH-T": 69.7
+      }
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "olmOCR2",
+      "overall": 70.4,
+      "digital": 79.9,
+      "photo": 67.2,
+      "latin": 76.7,
+      "non_latin": 63.3,
+      "private": 76.1,
+      "languages": {
+        "DE": 75.7,
+        "EN": 77.3,
+        "ES": 72.5,
+        "FR": 68.9,
+        "ID": 70.6,
+        "IT": 81.0,
+        "NL": 72.0,
+        "PT": 88.0,
+        "VI": 84.0,
+        "AR": 59.0,
+        "HI": 60.8,
+        "JP": 59.4,
+        "KO": 70.6,
+        "RU": 65.8,
+        "TH": 59.2,
+        "ZH": 68.6,
+        "ZH-T": 63.4
+      }
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "PaddleOCR-VL",
+      "overall": 69.6,
+      "digital": 87.6,
+      "photo": 63.6,
+      "latin": 72.1,
+      "non_latin": 66.7,
+      "private": 70.9,
+      "languages": {
+        "DE": 78.2,
+        "EN": 79.3,
+        "ES": 62.9,
+        "FR": 66.0,
+        "ID": 77.4,
+        "IT": 78.4,
+        "NL": 67.9,
+        "PT": 72.0,
+        "VI": 66.6,
+        "AR": 65.8,
+        "HI": 68.4,
+        "JP": 59.9,
+        "KO": 77.8,
+        "RU": 56.9,
+        "TH": 57.8,
+        "ZH": 78.2,
+        "ZH-T": 68.5
+      }
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "HunyuanOCR",
+      "overall": 68.3,
+      "digital": 80.2,
+      "photo": 64.3,
+      "latin": 72.4,
+      "non_latin": 63.7,
+      "private": 68.6,
+      "languages": {
+        "DE": 75.0,
+        "EN": 73.1,
+        "ES": 63.0,
+        "FR": 66.1,
+        "ID": 69.9,
+        "IT": 80.3,
+        "NL": 61.4,
+        "PT": 81.9,
+        "VI": 80.6,
+        "AR": 68.3,
+        "HI": 73.1,
+        "JP": 55.6,
+        "KO": 68.9,
+        "RU": 52.2,
+        "TH": 60.7,
+        "ZH": 66.8,
+        "ZH-T": 64.2
+      }
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "GLM-OCR",
+      "overall": 67.3,
+      "digital": 77.9,
+      "photo": 63.7,
+      "latin": 78.7,
+      "non_latin": 54.3,
+      "private": 68.8,
+      "languages": {
+        "DE": 82.7,
+        "EN": 84.5,
+        "ES": 75.8,
+        "FR": 76.2,
+        "ID": 79.7,
+        "IT": 82.8,
+        "NL": 80.2,
+        "PT": 77.4,
+        "VI": 69.2,
+        "AR": 21.7,
+        "HI": 39.6,
+        "JP": 65.5,
+        "KO": 61.2,
+        "RU": 64.2,
+        "TH": 27.4,
+        "ZH": 78.5,
+        "ZH-T": 76.7
+      }
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "MonkeyOCRv1.5",
+      "overall": 65.0,
+      "digital": 84.3,
+      "photo": 58.6,
+      "latin": 67.4,
+      "non_latin": 62.4,
+      "private": 69.0,
+      "languages": {
+        "DE": 70.8,
+        "EN": 74.9,
+        "ES": 55.6,
+        "FR": 60.3,
+        "ID": 73.8,
+        "IT": 75.9,
+        "NL": 66.3,
+        "PT": 67.2,
+        "VI": 61.4,
+        "AR": 60.1,
+        "HI": 56.8,
+        "JP": 57.0,
+        "KO": 78.9,
+        "RU": 51.7,
+        "TH": 55.6,
+        "ZH": 74.8,
+        "ZH-T": 64.1
+      }
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "Nanonets-OCR2-3B",
+      "overall": 64.2,
+      "digital": 79.2,
+      "photo": 59.3,
+      "latin": 71.4,
+      "non_latin": 56.2,
+      "private": 67.6,
+      "languages": {
+        "DE": 76.7,
+        "EN": 76.4,
+        "ES": 61.8,
+        "FR": 66.1,
+        "ID": 68.4,
+        "IT": 78.5,
+        "NL": 74.1,
+        "PT": 74.2,
+        "VI": 66.0,
+        "AR": 60.2,
+        "HI": 59.2,
+        "JP": 52.1,
+        "KO": 54.7,
+        "RU": 45.5,
+        "TH": 44.6,
+        "ZH": 68.3,
+        "ZH-T": 65.1
+      },
+      "source_name": "Nanonets-ocr2-3B"
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "LightOnOCR-2-1B",
+      "overall": 63.9,
+      "digital": 80.2,
+      "photo": 58.5,
+      "latin": 73.7,
+      "non_latin": 52.8,
+      "private": 68.1,
+      "languages": {
+        "DE": 72.4,
+        "EN": 79.1,
+        "ES": 65.0,
+        "FR": 62.1,
+        "ID": 72.9,
+        "IT": 82.9,
+        "NL": 70.2,
+        "PT": 83.8,
+        "VI": 74.9,
+        "AR": 64.2,
+        "HI": 59.0,
+        "JP": 50.5,
+        "KO": 41.9,
+        "RU": 54.3,
+        "TH": 51.4,
+        "ZH": 46.6,
+        "ZH-T": 54.7
+      }
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "Nanonets-OCR-s",
+      "overall": 63.7,
+      "digital": 78.8,
+      "photo": 58.7,
+      "latin": 71.3,
+      "non_latin": 55.0,
+      "private": 66.6,
+      "languages": {
+        "DE": 75.1,
+        "EN": 78.5,
+        "ES": 61.2,
+        "FR": 62.5,
+        "ID": 70.3,
+        "IT": 81.0,
+        "NL": 69.6,
+        "PT": 75.9,
+        "VI": 67.5,
+        "AR": 59.5,
+        "HI": 61.8,
+        "JP": 55.9,
+        "KO": 51.2,
+        "RU": 43.5,
+        "TH": 39.5,
+        "ZH": 67.4,
+        "ZH-T": 61.5
+      }
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "FalconOCR",
+      "overall": 56.3,
+      "digital": 72.4,
+      "photo": 51.1,
+      "latin": 69.3,
+      "non_latin": 41.6,
+      "private": 58.5,
+      "languages": {
+        "DE": 72.4,
+        "EN": 75.0,
+        "ES": 60.9,
+        "FR": 61.8,
+        "ID": 69.6,
+        "IT": 74.7,
+        "NL": 71.6,
+        "PT": 70.3,
+        "VI": 67.8,
+        "AR": 37.9,
+        "HI": 61.3,
+        "JP": 39.6,
+        "KO": 29.6,
+        "RU": 54.0,
+        "TH": 24.8,
+        "ZH": 39.7,
+        "ZH-T": 45.6
+      }
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "MonkeyOCR-pro-3B",
+      "overall": 52.2,
+      "digital": 68.0,
+      "photo": 47.0,
+      "latin": 65.1,
+      "non_latin": 37.6,
+      "private": 53.6,
+      "languages": {
+        "DE": 71.7,
+        "EN": 77.9,
+        "ES": 55.9,
+        "FR": 62.1,
+        "ID": 66.2,
+        "IT": 74.5,
+        "NL": 66.3,
+        "PT": 71.1,
+        "VI": 40.2,
+        "AR": 4.6,
+        "HI": 4.2,
+        "JP": 55.2,
+        "KO": 60.5,
+        "RU": 42.6,
+        "TH": 9.1,
+        "ZH": 72.2,
+        "ZH-T": 52.4
+      }
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "DeepSeek-OCR",
+      "overall": 51.8,
+      "digital": 80.7,
+      "photo": 42.2,
+      "latin": 54.5,
+      "non_latin": 48.9,
+      "private": 54.5,
+      "languages": {
+        "DE": 55.0,
+        "EN": 58.3,
+        "ES": 44.1,
+        "FR": 43.2,
+        "ID": 60.9,
+        "IT": 69.3,
+        "NL": 52.4,
+        "PT": 53.0,
+        "VI": 54.1,
+        "AR": 56.9,
+        "HI": 52.2,
+        "JP": 49.1,
+        "KO": 28.2,
+        "RU": 36.2,
+        "TH": 49.4,
+        "ZH": 59.7,
+        "ZH-T": 59.2
+      }
+    },
+    {
+      "type": "Specialized VLM",
+      "name": "MinerU-2.5-VLM",
+      "overall": 46.3,
+      "digital": 61.9,
+      "photo": 40.8,
+      "latin": 63.0,
+      "non_latin": 27.4,
+      "private": 48.7,
+      "languages": {
+        "DE": 68.8,
+        "EN": 78.4,
+        "ES": 54.7,
+        "FR": 57.3,
+        "ID": 67.5,
+        "IT": 75.2,
+        "NL": 60.4,
+        "PT": 58.8,
+        "VI": 46.0,
+        "AR": 1.3,
+        "HI": 9.0,
+        "JP": 39.1,
+        "KO": 14.7,
+        "RU": 8.6,
+        "TH": 11.3,
+        "ZH": 72.9,
+        "ZH-T": 62.2
+      }
+    },
+    {
+      "type": "Pipeline Tool",
+      "name": "PP-StructureV3",
+      "overall": 45.4,
+      "digital": 56.2,
+      "photo": 41.7,
+      "latin": 59.8,
+      "non_latin": 28.9,
+      "private": 49.6,
+      "languages": {
+        "DE": 60.4,
+        "EN": 68.7,
+        "ES": 54.4,
+        "FR": 49.8,
+        "ID": 69.6,
+        "IT": 68.9,
+        "NL": 55.5,
+        "PT": 58.4,
+        "VI": 52.7,
+        "AR": 1.0,
+        "HI": 7.7,
+        "JP": 56.2,
+        "KO": 15.4,
+        "RU": 7.5,
+        "TH": 11.9,
+        "ZH": 72.2,
+        "ZH-T": 59.1
+      }
+    },
+    {
+      "type": "Pipeline Tool",
+      "name": "MinerU-2.5-pipeline",
+      "overall": 33.5,
+      "digital": 57.6,
+      "photo": 25.4,
+      "latin": 46.5,
+      "non_latin": 18.7,
+      "private": 36.2,
+      "languages": {
+        "DE": 54.3,
+        "EN": 58.3,
+        "ES": 38.4,
+        "FR": 43.6,
+        "ID": 51.9,
+        "IT": 56.5,
+        "NL": 43.9,
+        "PT": 44.0,
+        "VI": 27.6,
+        "AR": 1.2,
+        "HI": 5.3,
+        "JP": 24.5,
+        "KO": 6.8,
+        "RU": 4.2,
+        "TH": 6.4,
+        "ZH": 53.9,
+        "ZH-T": 47.2
+      }
+    }
+  ]
+}

model_metadata.js

@@ -0,0 +1,39 @@
+/*
+ * Display names and first-party destinations for the initial MDPBench results.
+ * `name` is the evaluation identifier preserved in leaderboard.json; `label`
+ * is what readers see. Entries using a project page (rather than a versioned
+ * checkpoint) retain their exact evaluation identifier as the label.
+ */
+const MODEL_METADATA = {
+  "gemini-3-pro-preview": { label: "Gemini 3 Pro Preview", url: "https://ai.google.dev/gemini-api/docs/models/gemini-3-pro-preview" },
+  "kimi-k2.5": { label: "Kimi K2.5", url: "https://www.kimi.com/" },
+  "doubao-seed-2-0-pro-260215": { label: "Doubao-Seed-2.0-Pro (260215)", url: "https://api.volcengine.com/api-explorer/?action=ListFoundationModelVersions&groupName=%E5%9F%BA%E7%A1%80%E6%A8%A1%E5%9E%8B&serviceCode=ark&version=2024-01-01" },
+  "claude-sonnet-4-6": { label: "Claude Sonnet 4.6", url: "https://docs.anthropic.com/en/docs/about-claude/models/overview" },
+  "gpt-5.2-2025-12-11": { label: "GPT-5.2 (2025-12-11)", url: "https://platform.openai.com/docs/models" },
+  "Qwen3-VL-8B-Instruct": { label: "Qwen3-VL-8B-Instruct", url: "https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct" },
+  "Qwen3.5-Instruct-9B": { label: "Qwen3.5-Instruct-9B", url: "https://qwen.ai/" },
+  "InternVL3_5-8B": { label: "InternVL3_5-8B", url: "https://huggingface.co/OpenGVLab/InternVL3_5-8B" },
+  "MonkeyOCRv2-B-Parsing": { label: "MonkeyOCRv2-B-Parsing", url: "https://github.com/Yuliang-Liu/MonkeyOCR" },
+  "MonkeyOCRv2-S-Parsing": { label: "MonkeyOCRv2-S-Parsing", url: "https://github.com/Yuliang-Liu/MonkeyOCR" },
+  "dots.mocr": { label: "dots.mocr", url: "https://huggingface.co/rednote-hilab/dots.mocr" },
+  "chandra-ocr-2": { label: "Chandra OCR 2", url: "https://huggingface.co/datalab-to/chandra-ocr-2" },
+  "PaddleOCR-VL-1.5": { label: "PaddleOCR-VL-1.5", url: "https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.5" },
+  "HunyuanOCR-1.5": { label: "HunyuanOCR-1.5", url: "https://huggingface.co/tencent/HunyuanOCR" },
+  "dots.ocr": { label: "dots.ocr", url: "https://huggingface.co/rednote-hilab/dots.ocr" },
+  "PaddleOCR-VL-1.6": { label: "PaddleOCR-VL-1.6", url: "https://github.com/PaddlePaddle/PaddleOCR/blob/main/docs/version3.x/pipeline_usage/PaddleOCR-VL.md" },
+  "MinerU-2.5-Pro-1.2B": { label: "MinerU2.5-Pro-1.2B", url: "https://github.com/opendatalab/MinerU" },
+  "olmOCR2": { label: "olmOCR 2", url: "https://huggingface.co/allenai/olmOCR-2-7B-1025-FP8" },
+  "PaddleOCR-VL": { label: "PaddleOCR-VL-0.9B", url: "https://github.com/PaddlePaddle/PaddleOCR" },
+  "HunyuanOCR": { label: "HunyuanOCR", url: "https://huggingface.co/tencent/HunyuanOCR" },
+  "GLM-OCR": { label: "GLM-OCR", url: "https://huggingface.co/zai-org/GLM-OCR" },
+  "MonkeyOCRv1.5": { label: "MonkeyOCR v1.5", url: "https://github.com/Yuliang-Liu/MonkeyOCR" },
+  "Nanonets-OCR2-3B": { label: "Nanonets-OCR2-3B", url: "https://huggingface.co/nanonets" },
+  "LightOnOCR-2-1B": { label: "LightOnOCR-2-1B", url: "https://huggingface.co/lightonai/LightOnOCR-2-1B" },
+  "Nanonets-OCR-s": { label: "Nanonets-OCR-s", url: "https://huggingface.co/nanonets/Nanonets-OCR-s" },
+  "FalconOCR": { label: "Falcon-OCR", url: "https://huggingface.co/tiiuae/Falcon-OCR" },
+  "MonkeyOCR-pro-3B": { label: "MonkeyOCR-pro-3B", url: "https://github.com/Yuliang-Liu/MonkeyOCR" },
+  "DeepSeek-OCR": { label: "DeepSeek-OCR", url: "https://github.com/deepseek-ai/DeepSeek-OCR" },
+  "MinerU-2.5-VLM": { label: "MinerU2.5-VLM", url: "https://github.com/opendatalab/MinerU" },
+  "PP-StructureV3": { label: "PP-StructureV3", url: "https://github.com/PaddlePaddle/PaddleOCR" },
+  "MinerU-2.5-pipeline": { label: "MinerU2.5-pipeline", url: "https://github.com/opendatalab/MinerU" }
+};

style.css

@@ -1,28 +1,16 @@
-body {
-	padding: 2rem;
-	font-family: -apple-system, BlinkMacSystemFont, "Arial", sans-serif;
-}
-
-h1 {
-	font-size: 16px;
-	margin-top: 0;
-}
-
-p {
-	color: rgb(107, 114, 128);
-	font-size: 15px;
-	margin-bottom: 10px;
-	margin-top: 5px;
-}
-
-.card {
-	max-width: 620px;
-	margin: 0 auto;
-	padding: 16px;
-	border: 1px solid lightgray;
-	border-radius: 16px;
-}
-
-.card p:last-child {
-	margin-bottom: 0;
-}
+:root { --ink: #18251e; --muted: #65716a; --line: #d9ddd4; --paper: #f7f7f2; --lime: #d7fa4a; --blue: #1269db; }
+* { box-sizing: border-box; }
+body { margin: 0; color: var(--ink); background: var(--paper); font-family: Manrope, Arial, sans-serif; }
+.shell { max-width: 1320px; margin: 0 auto; padding: 32px 42px 68px; }
+.hero { padding: 16px 0 54px; max-width: 880px; }
+.brand { display: inline-block; margin-bottom: 67px; color: var(--ink); font-family: "DM Mono", monospace; font-size: 19px; font-weight: 700; letter-spacing: -.08em; text-decoration: none; }.brand span { color: var(--blue); }
+.eyebrow { margin: 0 0 10px; color: var(--blue); font-family: "DM Mono", monospace; font-size: 11px; font-weight: 700; letter-spacing: .09em; text-transform: uppercase; }
+h1, h2, p { margin-top: 0; } h1 { margin-bottom: 20px; font-size: clamp(39px, 5vw, 70px); font-weight: 800; letter-spacing: -.065em; line-height: .99; } h1 em { color: var(--blue); font-style: normal; }
+.intro { max-width: 660px; color: var(--muted); font-size: 16px; line-height: 1.7; }.links { display: flex; gap: 24px; margin-top: 27px; }.links a { color: var(--ink); font-family: "DM Mono", monospace; font-size: 12px; font-weight: 700; text-underline-offset: 4px; }
+.at-a-glance { display: grid; grid-template-columns: repeat(4, 1fr); border-top: 1px solid var(--ink); border-bottom: 1px solid var(--ink); margin-bottom: 73px; }.at-a-glance div { padding: 20px 20px 18px 0; border-right: 1px solid var(--line); }.at-a-glance div+div { padding-left: 20px; }.at-a-glance div:last-child { border: 0; }.at-a-glance strong, .at-a-glance span { display: block; }.at-a-glance strong { font-size: 25px; letter-spacing: -.05em; }.at-a-glance span { margin-top: 3px; color: var(--muted); font-family: "DM Mono", monospace; font-size: 10px; text-transform: uppercase; }
+.board { background: white; border: 1px solid var(--ink); box-shadow: 7px 7px 0 var(--lime); padding: 27px; }.board-topline { display: flex; align-items: end; justify-content: space-between; gap: 16px; }.board h2, .methodology h2 { margin: 0; font-size: 31px; letter-spacing: -.055em; }.updated { margin: 0; color: var(--muted); font-family: "DM Mono", monospace; font-size: 10px; }
+.controls { display: flex; align-items: center; gap: 11px; margin: 27px 0 20px; }.expand, .model-name { cursor: pointer; font: inherit; }.search { margin-left: auto; }.search input, select { border: 1px solid var(--line); border-radius: 0; padding: 10px 12px; color: var(--ink); background: white; font: 12px "DM Mono", monospace; }.search input { width: 155px; }.select-label { display: flex; align-items: center; gap: 8px; color: var(--muted); font: 10px "DM Mono", monospace; text-transform: uppercase; }
+.table-wrap { overflow-x: auto; } table { width: 100%; min-width: 720px; border-collapse: collapse; text-align: left; } th { padding: 10px 8px; border-bottom: 1px solid var(--ink); color: var(--muted); font: 10px "DM Mono", monospace; text-transform: uppercase; } td { padding: 13px 8px; border-bottom: 1px solid #e7e9e5; font-size: 12px; } .rank { color: var(--muted); font-family: "DM Mono", monospace; }.model-name { padding: 0; border: 0; background: none; color: var(--ink); font-size: 12px; font-weight: 800; text-align: left; }.score { font-family: "DM Mono", monospace; }.featured { color: var(--blue); font-weight: 800; }.type { display: inline-block; color: var(--muted); font-family: "DM Mono", monospace; font-size: 9px; }.type-general-vlm { color: #7a4b00; }.type-specialized-vlm { color: #1269db; }.type-pipeline-tool { color: #72766f; }.expand { width: 24px; height: 24px; border: 1px solid var(--line); background: white; color: var(--blue); font-size: 17px; line-height: 1; }.detail-row { display: none; background: #f3f8eb; }.detail-row.show { display: table-row; }.detail-row td { padding: 17px 20px 20px; }.details { display: flex; justify-content: space-between; gap: 28px; }.detail-title { margin: 15px 0 10px; font: 10px "DM Mono", monospace; text-transform: uppercase; }.detail-title:first-child { margin-top: 0; }.group-scores { display: flex; gap: 22px; }.group-scores span { display: flex; gap: 6px; font: 11px "DM Mono", monospace; }.group-scores b { color: var(--muted); font-weight: 400; }.language-grid { display: grid; grid-template-columns: repeat(9, 1fr); gap: 7px 18px; }.language-grid span { display: flex; gap: 6px; font: 11px "DM Mono", monospace; }.language-grid b { color: var(--muted); font-weight: 400; }.source-note { margin: 0; color: var(--muted); font-size: 11px; }.source-note code { color: var(--ink); }.methodology { display: grid; grid-template-columns: 1fr 1.3fr; gap: 90px; padding: 80px 0 4px; }.method-copy { color: var(--muted); font-size: 14px; line-height: 1.7; }.method-copy p+ p { margin-top: 16px; }footer { padding: 24px 42px; border-top: 1px solid var(--line); color: var(--muted); font: 10px "DM Mono", monospace; }.sr-only { position: absolute; width: 1px; height: 1px; overflow: hidden; clip: rect(0,0,0,0); }
+.rank-tabs { display: flex; background: #edf0ea; padding: 3px; }.rank-tabs button { border: 0; background: transparent; padding: 9px 12px; color: var(--muted); cursor: pointer; font: 700 11px "DM Mono", monospace; }.rank-tabs button.active { background: var(--ink); color: white; }.private { font-weight: 700; } table { min-width: 940px; }
+.model-link { display: inline-block; margin-left: 7px; color: var(--blue); font-family: "DM Mono", monospace; font-size: 12px; font-weight: 800; text-decoration: none; }.model-link:hover { text-decoration: underline; }
+@media (max-width: 720px) { .shell { padding: 22px 18px 44px; }.brand { margin-bottom: 46px; }.hero { padding-bottom: 38px; }.at-a-glance { grid-template-columns: repeat(2, 1fr); margin-bottom: 45px; }.at-a-glance div:nth-child(2) { border-right: 0; }.at-a-glance div:nth-child(n+3) { border-top: 1px solid var(--line); }.board { padding: 17px; box-shadow: 4px 4px 0 var(--lime); }.board-topline, .controls, .methodology { display: block; }.updated { margin-top: 10px; }.controls > * { margin: 10px 0 0; }.search { display: block; }.search input { width: 100%; }.select-label { justify-content: space-between; }.methodology { padding-top: 54px; }.method-copy { margin-top: 22px; }.language-grid { grid-template-columns: repeat(3, 1fr); }.details { display: block; }.source-note { margin-top: 16px; } footer { padding: 20px 18px; } }

tools/sync_leaderboard.py

@@ -0,0 +1,193 @@
+#!/usr/bin/env python3
+"""Synchronize the static leaderboard data from MDPBench's README table.
+
+The README is the source of truth for the initial official results. This tool
+parses its HTML table without third-party dependencies and preserves the stable
+evaluation identifiers already present in leaderboard.json.
+
+Examples:
+    python tools/sync_leaderboard.py --check
+    python tools/sync_leaderboard.py --write
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import re
+import sys
+from datetime import date
+from html.parser import HTMLParser
+from pathlib import Path
+from typing import Any
+
+
+LANGUAGES = [
+    "DE", "EN", "ES", "FR", "ID", "IT", "NL", "PT", "VI",
+    "AR", "HI", "JP", "KO", "RU", "TH", "ZH", "ZH-T",
+]
+SCORE_KEYS = ["overall", "digital", "photo", "latin", "non_latin", "private"]
+
+
+class ResultsTableParser(HTMLParser):
+    """Collect table rows and ``rowspan`` values from an HTML README table."""
+
+    def __init__(self) -> None:
+        super().__init__()
+        self.rows: list[list[tuple[str, int]]] = []
+        self._row: list[tuple[str, int]] | None = None
+        self._cell_text: list[str] | None = None
+        self._cell_rowspan = 1
+
+    def handle_starttag(self, tag: str, attrs: list[tuple[str, str | None]]) -> None:
+        if tag == "tr":
+            self._row = []
+        elif tag in {"td", "th"} and self._row is not None:
+            self._cell_text = []
+            attrs_dict = dict(attrs)
+            self._cell_rowspan = int(attrs_dict.get("rowspan") or 1)
+
+    def handle_data(self, data: str) -> None:
+        if self._cell_text is not None:
+            self._cell_text.append(data)
+
+    def handle_endtag(self, tag: str) -> None:
+        if tag in {"td", "th"} and self._cell_text is not None and self._row is not None:
+            text = " ".join("".join(self._cell_text).split())
+            self._row.append((text, self._cell_rowspan))
+            self._cell_text = None
+        elif tag == "tr" and self._row:
+            self.rows.append(self._row)
+            self._row = None
+
+
+def parse_number(value: str, *, model_name: str) -> float:
+    try:
+        return float(value)
+    except ValueError as exc:
+        raise ValueError(f"{model_name}: expected a numeric score, got {value!r}") from exc
+
+
+def parse_results(readme: Path) -> list[dict[str, Any]]:
+    parser = ResultsTableParser()
+    parser.feed(readme.read_text(encoding="utf-8"))
+
+    current_type: str | None = None
+    parsed: list[dict[str, Any]] = []
+    for row in parser.rows:
+        cells = [text for text, _ in row]
+        if not cells or cells[0] in {"Model Type", "All"}:
+            continue
+
+        # A category row has model type, model name, and 23 scores. Subsequent
+        # rows have the model name and the same 23 scores.
+        if len(cells) == 25:
+            current_type = re.sub(r"\s+", " ", cells[0])
+            model_name, score_values = cells[1], cells[2:]
+        elif len(cells) == 24 and current_type is not None:
+            model_name, score_values = cells[0], cells[1:]
+        else:
+            continue
+
+        if len(score_values) != 23:
+            raise ValueError(
+                f"{model_name}: expected 23 score columns from the README table, "
+                f"found {len(score_values)}"
+            )
+
+        numeric = [parse_number(value, model_name=model_name) for value in score_values]
+        latin_languages = numeric[4:13]
+        non_latin_languages = numeric[14:22]
+        parsed.append(
+            {
+                "type": current_type,
+                "source_name": model_name,
+                "overall": numeric[0],
+                "digital": numeric[1],
+                "photo": numeric[2],
+                "latin": numeric[3],
+                "non_latin": numeric[13],
+                "private": numeric[22],
+                "languages": dict(zip(LANGUAGES, latin_languages + non_latin_languages, strict=True)),
+            }
+        )
+
+    if not parsed:
+        raise ValueError("No Main Results rows were found in the README.")
+    return parsed
+
+
+def type_label(value: str) -> str:
+    compact = value.replace(" ", "")
+    return {
+        "GeneralVLMs": "General VLM",
+        "SpecializedVLMs": "Specialized VLM",
+        "PipelineTools": "Pipeline Tool",
+    }[compact]
+
+
+def build_document(source_rows: list[dict[str, Any]], existing: dict[str, Any]) -> dict[str, Any]:
+    existing_models = existing.get("models", [])
+    identifiers = {
+        model.get("source_name", model["name"]): model["name"]
+        for model in existing_models
+    }
+    source_names = {model.get("source_name") for model in existing_models if model.get("source_name")}
+    models = []
+
+    for source in source_rows:
+        source_name = source["source_name"]
+        model: dict[str, Any] = {
+            "type": type_label(source["type"]),
+            "name": identifiers.get(source_name, source_name),
+            "overall": source["overall"],
+            "digital": source["digital"],
+            "photo": source["photo"],
+            "latin": source["latin"],
+            "non_latin": source["non_latin"],
+            "private": source["private"],
+            "languages": source["languages"],
+        }
+        if source_name in source_names:
+            model["source_name"] = source_name
+        models.append(model)
+
+    missing = set(identifiers) - {row["source_name"] for row in source_rows}
+    if missing:
+        raise ValueError("Models in leaderboard.json absent from README: " + ", ".join(sorted(missing)))
+
+    benchmark = dict(existing.get("benchmark", {}))
+    benchmark["updated"] = date.today().isoformat()
+    return {"benchmark": benchmark, "models": models}
+
+
+def main() -> int:
+    root = Path(__file__).resolve().parents[1]
+    parser = argparse.ArgumentParser(description="Sync leaderboard.json from the MDPBench README results table.")
+    parser.add_argument("--readme", type=Path, default=root.parent / "README.md")
+    parser.add_argument("--output", type=Path, default=root / "leaderboard.json")
+    action = parser.add_mutually_exclusive_group(required=True)
+    action.add_argument("--check", action="store_true", help="Fail if the generated data differs from the current JSON.")
+    action.add_argument("--write", action="store_true", help="Write the generated data to leaderboard.json.")
+    args = parser.parse_args()
+
+    existing = json.loads(args.output.read_text(encoding="utf-8"))
+    generated = build_document(parse_results(args.readme), existing)
+    current = json.dumps(existing, ensure_ascii=False, sort_keys=True)
+    candidate = json.dumps(generated, ensure_ascii=False, sort_keys=True)
+    changed = current != candidate
+
+    if args.check:
+        if changed:
+            print("Leaderboard JSON is out of sync with README. Run with --write to update it.", file=sys.stderr)
+            return 1
+        print(f"In sync: {len(generated['models'])} models, {len(LANGUAGES)} language scores each.")
+        return 0
+
+    args.output.write_text(json.dumps(generated, ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
+    print(f"Wrote {len(generated['models'])} models to {args.output}")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())