# MHTML benchmarks

MHTML benchmarks comparing OfficeIMO with MimeKit on web archive reading and writing, with separate Windows, Linux and macOS runs.

A comparison runs OfficeIMO and other .NET libraries on the same validated work, one machine and one operating system at a time; a suite that measures OfficeIMO alone is listed without being published. Lower is better.

1 of 2 comparisons have a published run.

### MHTML archive reading

The MimeKit lane also parses the root HTML with AngleSharp, and the catalog cannot name a combined MimeKit and AngleSharp lane yet.

- Windows · net10.0 · not measured
- Windows · net8.0 · not measured
- Linux · net10.0 · not measured
- Linux · net8.0 · not measured
- macOS · net10.0 · not measured
- macOS · net8.0 · not measured

No published run yet.

### MHTML archive writing

- Windows · net10.0 · standard · 2026-11-20
- Windows · net8.0 · not measured
- Linux · net10.0 · not measured
- Linux · net8.0 · not measured
- macOS · net10.0 · standard · 2026-11-19
- macOS · net8.0 · not measured

- Windows
- net10.0
- Standard run
- 2026-11-20

AMD Ryzen 9 9950X3D2 (32 threads) · Windows 11 (10.0.26300.9974) · .NET 10.0.15

Workload `Case=Default`

| Case | OfficeIMO | MimeKit |
| --- | --- | --- |
| Large | 2.11 ms fastest 95% CI 2.04 ms – 2.17 ms | 4.27 ms 2.0× 95% CI 4.14 ms – 4.54 ms |
| Normal | 193 µs fastest 95% CI 191 µs – 199 µs | 236 µs 1.22× 95% CI 229 µs – 242 µs |
| Small | 10.2 µs 1.56× 95% CI 10.1 µs – 10.4 µs | 6.5 µs fastest 95% CI 6.4 µs – 6.8 µs |

- **OfficeIMO is 1.17× faster than MimeKit** OfficeIMO allocates 3.0× less · geometric mean of 3 cases

- Times are medians; results whose confidence intervals overlap are reported as even.
- Runner: BenchmarkDotNet · median · 5 warmups · 15 iterations · 1 launch · sequential

- macOS
- net10.0
- Standard run
- 2026-11-19

Apple M5 Max (18 threads) · macOS 27.2.1 (26B131) · .NET 10.0.15

Workload `Case=Default`

| Case | OfficeIMO | MimeKit |
| --- | --- | --- |
| Large | 1.94 ms fastest 95% CI 1.93 ms – 1.94 ms | 4.80 ms 2.5× 95% CI 4.72 ms – 4.88 ms |
| Normal | 172 µs fastest 95% CI 172 µs – 173 µs | 212 µs 1.23× 95% CI 212 µs – 213 µs |
| Small | 8.6 µs fastest 95% CI 8.5 µs – 8.6 µs | 9.4 µs 1.10× 95% CI 9.3 µs – 9.5 µs |

- **OfficeIMO is 1.49× faster than MimeKit** OfficeIMO allocates 2.9× less · geometric mean of 3 cases

- Times are medians; results whose confidence intervals overlap are reported as even.
- Runner: BenchmarkDotNet · median · 5 warmups · 15 iterations · 1 launch · sequential

## What a published result means

- **Same work, checked output.** Every library processes the same input, and its output is validated before it is timed.
- **Enough samples.** At least 3 warmups and 10 measured samples per case, with a 95% confidence interval for the median.
- **Even when it is close.** Only the library with the lowest time in a case is marked fastest. A library whose confidence interval overlaps the fastest one's is shown as even rather than slower.
- **Every case stays.** Cases another library wins, failed cases and partial runs are kept with their reason.
- **Traceable runs.** Each run records a clean source commit, the operating system, CPU, runtime and SDK, and the exact version and hash of every compared package.
- **Licences respected.** Libraries whose licence does not allow publishing results are run only locally and shown as not published.

Platforms are measured separately and never averaged together. Measure on hardware close to yours before deciding.

[Run the benchmarks yourself →](/docs/capabilities/benchmarks/)

<!-- PowerForge markdown artifact sha256:E7FDE3FCF7274E90C8B52BD422B7C12968783415A62C7ABE78B726025C25AFBD -->
