chore(site): update benchmark figures to 2026-03-28 snapshot

- EdgeParse: overall=0.7811 (+0.6 pp), speed=0.007 s/doc
- Speed ratios updated: 83× faster than Docling (was 12×),
  49× faster than PyMuPDF4LLM (was 6.9×), 2× faster than ODL (was 1.5×)
- TEDS claim corrected: 73% better than OpenDataLoader (was 83%)
- benchmark.ts lastUpdated → 2026-03-28 with exact per-engine figures
- Hero.astro eyebrow + subtitle updated
- index.mdx description, tagline, FeatureGrid card updated
- results.mdx Key Takeaways speed claim updated

Author: raphaelmansuy

site/src/components/landing/Hero.astro

@@ -15,7 +15,7 @@ const installCmd = 'pip install edgeparse';
   <div class="hero-content">
     <div class="hero-eyebrow">
       <span class="eyebrow-badge">#1 Non-ML PDF Parser</span>
-      <span class="eyebrow-text">Leads the current benchmark · 12× faster than Docling · Zero dependencies</span>
+      <span class="eyebrow-text">Leads the current benchmark · 83× faster than Docling · Zero dependencies</span>
       <svg class="eyebrow-arrow" width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round"><path d="m9 18 6-6-6-6"/></svg>
     </div>
 
@@ -25,7 +25,7 @@ const installCmd = 'pip install edgeparse';
     </h1>
 
     <p class="hero-subtitle">
-      Best published benchmark score without ML. 12× faster than Docling and 1.5× faster than OpenDataLoader. Zero GPU, zero OCR, zero JVM — just a 15&nbsp;MB Rust binary with the best reported scores across reading order, tables, headings, paragraphs, text quality, and speed.
+      Best published benchmark score without ML. 83× faster than Docling and 2× faster than OpenDataLoader. Zero GPU, zero OCR, zero JVM — just a 15&nbsp;MB Rust binary with the best reported scores across reading order, tables, headings, paragraphs, text quality, and speed.
     </p>
 
     <div class="hero-actions">

site/src/content/docs/benchmark/results.mdx

@@ -35,7 +35,7 @@ import { benchmarkSnapshot, formatSpeed, getBenchmarkTool } from '../../../data/
 ## Key Takeaways
 
 - **Latest published snapshot:** updated {benchmarkSnapshot.lastUpdated} on {benchmarkSnapshot.hardware} across {benchmarkSnapshot.documentCount} documents
-- **EdgeParse is the fastest** — {formatSpeed(getBenchmarkTool('EdgeParse').speedSeconds)} per document, 12× faster than Docling
+- **EdgeParse is the fastest** — {formatSpeed(getBenchmarkTool('EdgeParse').speedSeconds)} per document, 83× faster than Docling
 - **Highest overall score** — {getBenchmarkTool('EdgeParse').overall.toFixed(3)} across the current six-engine comparison
 - **Best structure metrics** — leading NID ({getBenchmarkTool('EdgeParse').nid.toFixed(3)}), TEDS ({getBenchmarkTool('EdgeParse').teds.toFixed(3)}), and MHS ({getBenchmarkTool('EdgeParse').mhs.toFixed(3)})
 - **Best text metrics** — also leads paragraph boundaries, text quality, and table-detection F1 in the full benchmark report

site/src/content/docs/index.mdx

@@ -1,12 +1,12 @@
 ---
 title: EdgeParse — Fast PDF Parser. Zero ML. Best Benchmark Score.
-description: EdgeParse extracts structured Markdown, JSON, and HTML from born-digital PDFs. 0.787 overall and 0.064 s/doc on the current 200-document benchmark. Python, Node.js, Rust, CLI, WebAssembly. Zero GPU. Zero OCR.
+description: EdgeParse extracts structured Markdown, JSON, and HTML from born-digital PDFs. 0.781 overall and 0.007 s/doc on the current 200-document benchmark. Python, Node.js, Rust, CLI, WebAssembly. Zero GPU. Zero OCR.
 template: splash
 hero:
   title: |
     PDF parsing for <span class="font-black text-transparent bg-clip-text bg-gradient-to-b from-accent-700 to-accent-400">AI Agents</span>
   tagline: |
-    The PDF-to-Markdown engine that leads the current benchmark without ML. 12× faster than Docling · 1.5× faster than OpenDataLoader · best reported scores across reading order, tables, headings, paragraphs, text quality, and speed. Python · Node.js · WebAssembly · Rust · CLI.
+    The PDF-to-Markdown engine that leads the current benchmark without ML. 83× faster than Docling · 2× faster than OpenDataLoader · best reported scores across reading order, tables, headings, paragraphs, text quality, and speed. Python · Node.js · WebAssembly · Rust · CLI.
   actions:
     - text: Get Started Free
       link: /getting-started/quick-start-python/
@@ -139,8 +139,8 @@ const html = convert_to_string(bytes, 'html');
   title="Everything Your AI Stack Needs From a PDF"
   subtitle="EdgeParse is the only PDF parser with ML-level accuracy that runs without ML — in Python, Node.js, the browser, and Rust."
   features={[
-    { icon: 'zap', title: '12× Faster Than Docling', description: `${formatSpeed(getBenchmarkTool('EdgeParse').speedSeconds)} on ${benchmarkSnapshot.hardware}. 6.9× faster than PyMuPDF4LLM and 1.5× faster than OpenDataLoader. Parallel per-page processing via Rayon — CPU only.` },
-    { icon: 'table', title: 'Best-in-Class Table Extraction', description: `TEDS score of ${getBenchmarkTool('EdgeParse').teds.toFixed(3)} — best in the current published comparison and 83% better than OpenDataLoader heuristic mode (${getBenchmarkTool('OpenDataLoader').teds.toFixed(3)}). Ruling-line + borderless cluster detection with merged cell support.` },
+    { icon: 'zap', title: '83× Faster Than Docling', description: `${formatSpeed(getBenchmarkTool('EdgeParse').speedSeconds)} on ${benchmarkSnapshot.hardware}. 49× faster than PyMuPDF4LLM and 2× faster than OpenDataLoader. Parallel per-page processing via Rayon — CPU only.` },
+    { icon: 'table', title: 'Best-in-Class Table Extraction', description: `TEDS score of ${getBenchmarkTool('EdgeParse').teds.toFixed(3)} — best in the current published comparison and 73% better than OpenDataLoader heuristic mode (${getBenchmarkTool('OpenDataLoader').teds.toFixed(3)}). Ruling-line + borderless cluster detection with merged cell support.` },
     { icon: 'target', title: 'Multi-Column Reading Order', description: `XY-Cut++ reads multi-column layouts, sidebars, and mixed content in the correct logical order. NID score of ${getBenchmarkTool('EdgeParse').nid.toFixed(3)} — highest in the current benchmark snapshot.` },
     { icon: 'layers', title: 'Full Document Hierarchy', description: `Headings, paragraphs, lists, figures — all classified with nesting. MHS score of ${getBenchmarkTool('EdgeParse').mhs.toFixed(3)}, best among the compared engines in the current release snapshot.` },
     { icon: 'globe', title: 'WebAssembly: Runs in the Browser', description: 'The only PDF parser with a WebAssembly build. Full Rust engine in the browser — PDF data never leaves the device. No server, no uploads, offline-capable.' },

site/src/data/benchmark.ts

@@ -9,58 +9,58 @@ export interface BenchmarkTool {
 }
 
 export const benchmarkSnapshot = {
-  lastUpdated: "2026-03-26",
+  lastUpdated: "2026-03-28",
   hardware: "Apple M4 Max",
   documentCount: 200,
   tools: [
     {
       name: "EdgeParse",
-      nid: 0.889,
-      teds: 0.596,
-      mhs: 0.553,
-      overall: 0.787,
-      speedSeconds: 0.064,
+      nid: 0.8846,
+      teds: 0.5590,
+      mhs: 0.5540,
+      overall: 0.7811,
+      speedSeconds: 0.007,
       isHighlight: true,
     },
     {
       name: "Docling (IBM)",
-      nid: 0.867,
-      teds: 0.54,
-      mhs: 0.438,
-      overall: 0.745,
-      speedSeconds: 0.768,
+      nid: 0.8665,
+      teds: 0.5404,
+      mhs: 0.4384,
+      overall: 0.7452,
+      speedSeconds: 0.584,
     },
     {
       name: "OpenDataLoader",
-      nid: 0.873,
-      teds: 0.326,
-      mhs: 0.442,
-      overall: 0.733,
-      speedSeconds: 0.094,
+      nid: 0.8611,
+      teds: 0.3234,
+      mhs: 0.4360,
+      overall: 0.7233,
+      speedSeconds: 0.014,
     },
     {
       name: "PyMuPDF4LLM",
-      nid: 0.852,
-      teds: 0.323,
-      mhs: 0.407,
-      overall: 0.71,
-      speedSeconds: 0.439,
+      nid: 0.8522,
+      teds: 0.3233,
+      mhs: 0.4066,
+      overall: 0.7103,
+      speedSeconds: 0.327,
     },
     {
       name: "LiteParse",
-      nid: 0.815,
-      teds: 0,
-      mhs: 0.001,
-      overall: 0.564,
-      speedSeconds: 0.196,
+      nid: 0.8148,
+      teds: 0.0000,
+      mhs: 0.0012,
+      overall: 0.5642,
+      speedSeconds: 0.160,
     },
     {
       name: "MarkItDown",
-      nid: 0.808,
-      teds: 0.193,
-      mhs: 0.001,
-      overall: 0.564,
-      speedSeconds: 0.149,
+      nid: 0.8075,
+      teds: 0.1925,
+      mhs: 0.0012,
+      overall: 0.5639,
+      speedSeconds: 0.123,
     },
   ] satisfies BenchmarkTool[],
 };