Decoder comparison
Recorded run: dec-2026-10-30-amd (2026-10-30). CSV · Environment · Methodology and limits.
Each value is the median of C mean time / decoder mean time across all 8 datasets, equally weighted. Higher is faster; 1× matches C. Qualities are ordered by mbrotli median speed / C. Medians do not imply statistical significance.
All four decoders restore identical C-generated streams. Source quality is an encoder setting, not a decoder setting. Burli decodes q0–q11; SIMD Brotli shares Rust brotli’s decoder and is omitted. Compressed sizes and ratios are properties of those shared inputs, so there is no decoder size ranking.
| Source quality | Google C | mbrotli | Rust brotli | Burli | mbrotli / Burli |
|---|---|---|---|---|---|
| q4 | 1.000× | 3.461× | 0.574× | 3.407× | 1.003× |
| q2 | 1.000× | 3.325× | 0.570× | 3.362× | 1.056× |
| q1 | 1.000× | 3.196× | 0.445× | 3.129× | 1.020× |
| q0 | 1.000× | 3.059× | 0.446× | 2.873× | 1.036× |
| q11 | 1.000× | 2.847× | 0.584× | 2.662× | 1.067× |
| q10 | 1.000× | 2.770× | 0.581× | 2.632× | 1.051× |
| q7 | 1.000× | 2.574× | 0.562× | 2.511× | 0.982× |
| q5 | 1.000× | 2.572× | 0.574× | 2.555× | 0.982× |
| q6 | 1.000× | 2.568× | 0.569× | 2.545× | 0.982× |
| q9 | 1.000× | 2.553× | 0.569× | 2.536× | 0.986× |
| q3 | 1.000× | 2.548× | 0.563× | 2.536× | 0.986× |
| q8 | 1.000× | 2.547× | 0.572× | 2.533× | 0.988× |
The last column takes the median of direct Burli time / mbrotli time ratios; it does not divide the two medians relative to C.
Open a source quality for all datasets, compressed/restored byte counts, mean latency, confidence bounds and throughput. Measurements include cold construction, allocation, decoding and disposal. C receives the known output capacity; Rust brotli includes its native 4 KiB I/O adapter. See the linked methodology for reproducible commands and the limits of this single-host comparison.