# video-hosting-bench – round 1 (2026-10-12 to 2026-10-12)

## Storage and traffic per minute (with cap)

| rung | MB/min | Mbit/s | GB/h |
|---|---|---|---|
| 360p | 4.18 | 0.56 | 0.25 |
| 540p | 8.17 | 1.09 | 0.49 |
| 720p | 12.80 | 1.71 | 0.77 |
| 1080p | 25.43 | 3.39 | 1.53 |

## Costs per month (EUR, net)

| provider | S | M | L |
|---|---|---|---|
| YouTube | 0.00 | 0.00 | 0.00 |
| Vimeo (Einzelpersonen) | Standard 22.36 | no plan fits (videos a year, one seat) | no plan fits (videos a year, one seat) |
| Vimeo Studio | 357.75 | 357.75 | no plan fits (bandwidth) |
| Bunny Stream | 0.93 | 9.29 | 92.92 |
| Bunny Stream (Volume) | 0.89 | 4.99 | 49.85 |
| Cloudflare Stream | 11.18 | 76.02 | 737.86 |
| Eigener Server | 20.56 | 24.34 | 63.40 |
| Eigener Server + CDN | 21.46 | 32.95 | 149.54 |

Model at equal assumed byte volumes, using our H.264 encodes for Bunny storage/delivery and Vimeo delivery; provider encodes and equivalent visual quality were not measured. No video preload before playback, extra buffering/repeated delivery, labour, original archive, backups or spare storage are priced. Cloudflare delivery uses continuous minutes without per-segment rounding. Storage is a monthly snapshot, not a growing yearly inventory. Studio includes team features and support not replaced by a bare server.

### Sensitivity: all delivery at 1080p (EUR/month)

| provider | S | M | L |
|---|---|---|---|
| Bunny Stream (Volume) (all 1080p) | 0.92 | 9.21 | 92.07 |
| Eigener Server (all 1080p) | 20.56 | 24.34 | 63.40 |

Peak viewers (mean concurrency x peak factor): S 2, M 17, L 174.

Provisioned Hetzner Volume (minimum 10 GB, rounded up to whole GB, no spare capacity): S 10 GB, M 76 GB, L 759 GB.

## Volumes per scenario (for recalculating)

| scenario | MB per delivered minute (mix) | traffic GB/month | storage GB (all rungs) | uploads in the first year | originals GB (project) | originals GB (test clips) | re-encode h (4 CPUs) |
|---|---|---|---|---|---|---|---|
| S | 12.84 | 96 | 7.6 | 80 | 1.7 | 9.7 | 2.0 |
| M | 12.84 | 963 | 75.9 | 800 | 17.0 | 97.4 | 19.6 |
| L | 12.84 | 9,631 | 758.5 | 8,000 | 170.5 | 973.8 | 195.9 |

Traffic at the scenario's views with 50% watched; uploads in the first year = inventory + 12 x 5% of it; originals at 11.4 MB/min (project) and 64.9 MB/min (test clips); re-encoding at 0.78 x real time on 4 CPUs.

## Break-even (scenario M, views per month)

| from | views (half watched) | a quarter watched | all watched |
|---|---|---|---|
| own server cheaper than Cloudflare Stream | 11,000 | 23,000 | 6,000 |
| own server cheaper than Bunny Stream | 137,000 | 275,000 | 69,000 |
| own server cheaper than Bunny Stream (Volume Network) | 275,000 | 549,000 | 137,000 |
| own server cheaper than Vimeo (plans for individuals) | never in the range | never in the range | never in the range |
| own server cheaper than Vimeo Studio (companies) | 1,000 | 1,000 | 1,000 |
| no Vimeo plan for individuals fits (uploads a year, storage or 2 TB a month) | 1,000 | 1,000 | 1,000 |
| Vimeo Studio does not fit (36 TB a year) | 156,000 | 311,000 | 78,000 |
| model exceeds measured lab capacity (not a production capacity guarantee) | 9,216,000 | never in the range | 4,608,000 |

## Embedded players (median of 5 Lighthouse runs; capture before and after the click)

| variant | kB before | third-party hosts before | cookies before | new hosts after | playing | main-thread TBT ms (mobile emulation) |
|---|---|---|---|---|---|---|
| baseline | 1 | 0 | 0 | 0 | False | 0 |
| own | 165 | 0 | 0 | 0 | True | 228 |
| youtube_facade | 105 | 1 | 0 | 7 | True | 0 |
| youtube_nocookie | 1244 | 7 | 0 | 1 | True | 0 |
| youtube | 1235 | 9 | 5 | 1 | True | 0 |
| vimeo | 548 | 4 | 3 | 0 | False | 0 |
| bunny | 4471 | 17 | 0 | 11 | True | 0 |
| cloudflare | 411 | 3 | 0 | 0 | True | 0 |

LCP and Lighthouse aggregate scores are not compared: Lighthouse evaluates the main frame only, and the iframe variants varied between passes (DEVIATIONS.md, 2026-10-13). TBT describes main-thread work in this setup, not player performance overall; raw values stay in summary.json.

## Bitrate cap (top rung / original)

| case | n | median | q1–q3 | above the original |
|---|---|---|---|---|
| test clips, cap | 13 | 0.402 | 0.271–0.523 | 0% |
| test clips, no cap | 13 | 0.402 | 0.271–0.523 | 0% |
| project, all, cap | 546 | 0.771 | 0.674–0.869 | 1% |
| project, sample, cap | 60 | 0.898 | 0.832–0.926 | 2% |
| project, sample, no cap | 60 | 1.059 | 0.939–1.173 | 63% |

Project: the cap applies to 212 of 546 videos (39%); mirror of 2026-09-24.

## Load test

| type | EUR/month | viewers (mix) | Gbit/s | p95 ms | viewers (all 1080p) | Gbit/s |
|---|---|---|---|---|---|---|
| cpx22 | 19.99 | 2400 | 3.99 | 45 | ≥ 1600 | 5.43 |
| cpx32 | 35.99 | ≥ 3200 | 5.43 | 14 | ≥ 1600 | 5.35 |

≥: no other target carried the stage where this series broke down, so the cause (the server or the test set-up, e.g. the generator) is not identified; the value is a lower bound. Network check (iperf3, 8 streams, 20 s, without TLS): cpx22 to ccx33 13.5 Gbit/s, cpx32 + cpx22 to ccx33 13.0 Gbit/s, cpx32 to ccx33 12.7 Gbit/s.