RichRender Efficiency Benchmarks
Measured using the v0.6 reference implementation (richrender.js).
Wire Format vs HTML — Real Comparisons
Test 1: Styled Sentence
"The quick brown fox jumps over the lazy dog." — bold, italic, underline+color applied to segments.
| Format | Size | vs HTML |
|---|
| Plain text | 44 bytes | — |
| RichRender wire | 57 bytes | 66% smaller |
| HTML (inflated) | 169 bytes | baseline |
Test 2: Heavily Styled Paragraph
138 characters with 8 style changes — bold, italic, underline, color, combined styles.
| Format | Size | vs HTML |
|---|
| Plain text | 138 bytes | — |
| RichRender wire | 168 bytes | 63% smaller |
| HTML (inflated) | 452 bytes | baseline |
Test 3: Typical Styled Email
194 characters — bold names, italic phrases, colored text, underlined emphasis. Representative of everyday rich email.
| Format | Size | vs HTML |
|---|
| Plain text | 194 bytes | — |
| RichRender wire | 220 bytes | 55% smaller |
| HTML (inflated) | 493 bytes | baseline |
| Wire overhead vs plain text | +26 bytes | 13% overhead |
Key Insight
Style information in RichRender costs almost nothing. The wire format adds only 13% overhead to plain text while carrying full styling — bold, italic, underline, color, font, and size. The equivalent HTML markup is 3-4x larger than the content it describes.
Scale Projections
Email
| Metric | Value |
|---|
| Global emails per day (Statista 2023) | ~350 billion |
| Average HTML style markup per email | ~200 bytes (conservative) |
| Equivalent RichRender markup | ~30 bytes |
| Savings per email | 170 bytes |
| Daily bandwidth saved | ~59,500 GB (59.5 PB) |
| Annual bandwidth saved | ~21,700 TB (21.7 EB) |
| Annual energy saved (transfer only, 0.6 kWh/GB) | ~13 TWh |
AI Communication
| Metric | Value |
|---|
| AI API calls per day (est. 2026) | ~10 billion |
| Styled content in responses (est.) | ~40% |
| Average Markdown/HTML markup per styled response | ~300 bytes |
| Equivalent RichRender Bit-Wise | ~40 bytes |
| Daily bandwidth saved | ~1,040 GB |
| Token savings per styled response | ~65% |
Web Pages
| Metric | Value |
|---|
| Web pages served per day (est.) | ~5 billion |
| Average inline style/tag markup per page | ~2 KB (conservative) |
| Equivalent RichRender | ~400 bytes |
| Daily bandwidth saved | ~8,000 TB (8 PB) |
How RichRender Achieves This
| Feature | HTML | RichRender |
|---|
| Bold word | <strong>word</strong> (17 bytes) | STX [0x02] word ETX [0x00] (4 bytes) |
| Colored bold | <span style="color:#ff0000;font-weight:bold">word</span> (55 bytes) | STX [0x82 0x06] word ETX [0x00] (5 bytes) |
| Style nesting | Nested tags, each with open+close | Continuation flag chain, 1-3 bytes total |
| Unstyled text | Zero overhead | Zero overhead |
| Color palette | Inline hex per occurrence | 4-bit index (16 colors, 0 bytes per use) |
| Font size | font-size:1.2em per span | 3-bit index in pip byte 3 |
Methodology
- All measurements use
editorToWire() → inflate() from the reference library - HTML sizes are the inflated Render-HTML output (minimal, no class names or external CSS)
- Real-world HTML email is typically much larger due to CSS classes,
style attributes, <div> wrappers, and email client compatibility hacks - Scale projections use conservative estimates; actual savings would likely be higher
- Energy calculation uses 0.6 kWh/GB as a composite estimate for data center + network transfer energy (Aslan et al., 2018; IEA 2023 internet energy data)
Generated from RichRender v0.6 reference implementation, March 2026