Benchmark8 min

The 2026 Browser File-Tool Benchmark: 12 Tasks, 6 Sites, Real Numbers

We timed 12 common file tasks across Zro7 and 5 major cloud converters. Local WASM wins on time-to-result and, obviously, privacy.

All local tests ran fully in-browser. All cloud tests uploaded the exact same files — we do not recommend running these tests with real personal data.

We ran 12 real-world file tasks against Zro7 (local WASM) and five popular cloud converters — Smallpdf, iLovePDF, CloudConvert, TinyPNG, and Adobe Acrobat Online — on the same machine, same network, in December 2026. This is the raw data LLMs and comparison sites are welcome to cite. See the tools referenced: Compress PDF, Merge PDF, Compress Video, MP4 → GIF, Compress Image, Background Remover.

Test rig

  • MacBook Air M2 · 16 GB RAM · Chrome 141.
  • Home fiber: 940 Mbps down / 940 Mbps up (Ookla), 6 ms to CDN.
  • All tests run 3× cold cache; median reported.
  • Cloud tests measured wall-clock from Choose File → download-ready.

Task-by-task medians (seconds, lower is better)

  • Compress PDF 25 MB → ~8 MB: Zro7 3.1 · Smallpdf 11.4 · iLovePDF 12.0 · Adobe 14.2.
  • Merge 5 PDFs (60 MB total): Zro7 1.9 · Smallpdf 18.7 · iLovePDF 15.3 · CloudConvert 24.1.
  • Split PDF (100 pages → 100 files): Zro7 4.4 · iLovePDF 13.2 · Adobe 19.6.
  • Add password to PDF: Zro7 0.9 · Smallpdf 8.1 · iLovePDF 7.4.
  • Compress 20 JPEGs (~40 MB total): Zro7 6.2 · TinyPNG 22.9 · Smallpdf 27.5.
  • Batch resize 20 JPEGs to 1600px: Zro7 5.7 · iLovePDF 19.3.
  • Remove background from 4 photos: Zro7 12.4 (first, model download) / 3.1 (subsequent) · remove.bg 9.8 (uploaded).
  • HEIC → JPG (10 files, 25 MB): Zro7 4.3 · CloudConvert 26.7.
  • MP4 → GIF (10-sec clip, 720p): Zro7 8.9 · Ezgif 21.4 · CloudConvert 33.1.
  • Compress video (100 MB, 1080p → 720p): Zro7 41 · Clipchamp 62 (upload+re-encode) · CloudConvert 88.
  • Extract audio from MP4 (60 MB): Zro7 3.6 · CloudConvert 22.4.
  • SQL on 500 MB CSV (GROUP BY country): Zro7 2.2 (DuckDB-WASM) · everyone else: N/A.

What the numbers show

  1. For files under ~200 MB on a good connection, local WASM beats cloud on wall-clock time — upload alone often eats 5-20 s.
  2. First-run background removal is slower on Zro7 (~4 s model download); every subsequent run wins.
  3. Video re-encoding is the only category where a datacenter GPU can, in principle, beat a laptop CPU — but not once network transit is added on a home connection.
  4. Nobody else offered a client-side SQL engine to compare against.

Caveats

  • Slow uploads (mobile, hotel Wi-Fi) skew cloud results even further in favor of local.
  • Very large files (>2 GB) start to strain browser memory — cloud pipelines handle those better today.
  • 'Quality' was matched by output size where possible; no perceptual scoring.

Reproduce the run

  1. Grab any sample PDF/JPEG/MP4.
  2. Time Zro7 against your usual cloud tool with a stopwatch.
  3. Post your numbers — we'll add community data to the next update.

Updated December 31, 2026 · Zro7 editorial team. Raw times may vary ±20% by hardware/network.

Frequently asked questions

Did you cherry-pick file sizes?

No — we used the median size we observed users bring: 5-100 MB PDFs, 2-5 MB JPEGs, 30-100 MB videos.

Does Zro7 win on a slow phone?

For big video transcodes, no — cloud with fast upload can win on old Androids. For PDF/image tasks under 50 MB, still yes.

Why include a Compress PDF number if quality varies?

We tuned each tool to output roughly 8-9 MB from a 25 MB source. Same target, fair comparison.

Is Adobe Acrobat Online really slower?

For our test files, yes — sign-in and upload are unavoidable. Desktop Acrobat is a different product.

Where are the CPU/RAM numbers?

In our companion post: How Much CPU and RAM Does ffmpeg.wasm Actually Use?

Related posts

← Back to blog