Proxy client benchmark
Up to eight Mac apps sent the same traffic through the same servers, one app at a time, on battery. Three setups, measured separately: pick one to see its leaderboard and every measure behind it.
Score = 35% efficiency under traffic + 35% efficiency at idle + 30% performance; lower is better. Numbers are medians over all rounds; app power and energy are macOS's own per-process estimates for the tunnel process.
Chain + rules: 4 apps, 6 rounds
The everyday setup: the app chains two proxies, and rules decide which sites use the tunnel.
| App | Score | Idle wakeups per s | Avg idle power µW | Power at 100 Mbit/s mW | Energy at 100 Mbit/s J per GB | CPU at 100 Mbit/s % | Full speed Mbit/s | First byte ms |
|---|---|---|---|---|---|---|---|---|
| Egern 2.20.0 | 1.61 | 4.07 | 13.6 | 11.6 | 0.932 | 17.6 | 181 | 465 |
| SFM 1.15.0-alpha.12 | 2.20 | 15.7 | 16.5 | 12.6 | 1.01 | 22.1 | 182 | 465 |
| Anywhere 2.4 | 2.60 | 0.462 | 50.6 | 51.8 | 4.14 | 29.9 | 182 | 471 |
| Shadowrocket 2.2.92 | 3.58 | 4.08 | 20.8 | 50.2 | 4.01 | 50.9 | 179 | 541 |
iPhone check (16 Oct): on an iPhone 11 Pro (iOS 27, same chain + rules, 2 rounds, whole-phone power from the battery gauge over USB), Anywhere 2.4 and Egern tied: 0.93 W vs 0.93–0.95 W at a steady 50 Mbit/s download (about 148–152 J per GB for the whole phone), the same single-connection speed (70–75 Mbit/s) and page first-byte times within run-to-run noise. Idle draw was within noise of no VPN for both. The Mac's ~4× gap in app energy per GB is in the app's own small share of power and doesn't show at the whole-device level.
Basic: 8 apps, 4 rounds
The simplest setup: one proxy, no rules, everything through the tunnel.
| App | Score | Idle wakeups per s | Avg idle power µW | Power at 100 Mbit/s mW | Energy at 100 Mbit/s J per GB | CPU at 100 Mbit/s % | First byte ms |
|---|---|---|---|---|---|---|---|
| Egern 2.20.0 | 2.63 | 4.15 | 20.8 | 3.82 | 0.306 | 8.12 | 462 |
| Anywhere 2.3 | 2.90 | 0.258 | 12.4 | 11.8 | 0.947 | 23.9 | 481 |
| SFM 1.15.0-alpha.12 | 3.76 | 12.4 | 18.4 | 7.08 | 0.567 | 14.1 | 464 |
| Karing 1.2.25 | 3.83 | 9.18 | 16.5 | 12.5 | 0.997 | 23.4 | 469 |
| Shadowrocket 2.2.92 | 4.13 | 1.8 | 12.8 | 16.8 | 1.35 | 30.6 | 544 |
| Rabbit Hole 1.6.0 | 5.27 | 14.7 | 22.7 | 28.1 | 2.25 | 52.4 | 463 |
| V2Box 10.1.9 | 6.60 | 51.4 | 117 | 26.3 | 2.11 | 37.2 | 688 |
| Happ 6.1.0 | 6.88 | 83.3 | 411 | 43 | 3.44 | 56.8 | 684 |
Top speed: 8 apps, 1 round
A lab test on the Mac itself, with no internet in the way. One round only.
| App | Top upload Gbit/s | Power at 500 Mbit/s mW | Energy at 500 Mbit/s J per GB | CPU at 500 Mbit/s % | Idle wakeups per s | Avg idle power µW |
|---|---|---|---|---|---|---|
| SFM 1.15.0-alpha.12 | 10.2 | 41.2 | 0.662 | 38.4 | 34 | 202 |
| Karing 1.2.25 | 4.81 | 56.5 | 0.905 | 85.6 | 50.9 | 642 |
| Egern 2.20.0 | 3.46 | 18.3 | 0.293 | 24 | 4.17 | 25.9 |
| Rabbit Hole 1.6.0 | 1.91 | 116 | 1.86 | 129 | 22.1 | 41.9 |
| Shadowrocket 2.2.92 | 1.47 | 67 | 1.07 | 69.6 | 1.6 | 11.5 |
| Anywhere 2.3 | 0.953 | 191 | 3.07 | 106 | 2.37 | 585 |
| Anywhere 2.4* | 5.8 | 154 | 2.48 | 52.3 | – | – |
| V2Box 10.1.9 | 0.393 | – | – | – | 62.3 | 222 |
| Happ 6.0.0 | 0.25 | – | – | – | 75.2 | 425 |
* Anywhere 2.4 was rechecked alone on 17 Oct after the update: same lab and Mac, 1 round, with a profiler attached (adds a little CPU). Top download 4.1 Gbit/s. Idle was not measured.
Footnote: this is not a professional lab test, just a personal, nerdy-interest one: one Mac on Wi-Fi, a few days, a handful of rounds. On a phone over cellular the results may differ.
Scroll back through thirty years of tunnels
Each technique sits where it first shipped, newest on the left. Under the read line you get its 2026 status in four censored countries, how it connects, and an estimate of its battery cost.
What to run in 2026, and the path it takes
Pick your situation and compare the top setups. Each one shows who sees what on the way from your device to the website.
Every figure links to its source on the sources page.