2026 performance inequality update: budgets grow but median pages now outweigh DOOM

Updated baselines for 2026

The annual "performance inequality gap" analysis published on Infrequently.org, recently surfaced on the Hacker News front page, has refreshed the reference conditions web developers should test against. For 2026, the network profile is a 9 Mbps downlink, 3 Mbps uplink and 100 ms round-trip time. The recommended reference devices are the Samsung Galaxy A24 4G phone and the HP 14 laptop.

The pairing is about calibration: these settings emulate the 75th-percentile user experience, which means a full quarter of visitors will be on slower hardware or worse connections than your test setup.

What the budgets allow

Feeding those parameters into the post's revised budget calculator yields critical-path byte thresholds for three- and five-second page loads. The analysis distinguishes two build styles: JS-light pages, where JavaScript is 15% of critical-path bytes, and JS-heavy pages, where it is half.

For a three-second load, JS-light pages get 2.0 MiB total (0.3 MiB of JavaScript, 1.7 MiB of everything else), while JS-heavy pages get just 1.2 MiB, split evenly. At five seconds the figures are 3.7 MiB (0.57 MiB JS) versus 2.3 MiB (1.15 MiB JS).

Two caveats stand out. First, the budgets assume two TLS connections; opening four early connections costs 350 KiB at three seconds (down to 1.5 MiB and 935 KiB respectively) and nearly half a megabyte at five seconds, which the post frames as an argument for HTTP/2 or HTTP/3 and consolidating connections. Second, JS-heavy pages receive less total budget for the same time target because JavaScript carries a higher CPU cost per byte, what the author calls a coffin-corner effect, explaining why image- and CSS-heavy sites deliver more byte-for-byte.

Pages keep outgrowing the gains

There is genuinely good news: budgets are expanding. Compared with early 2024 estimates, the three-second budget grew by more than 600 KiB and the five-second budget by a full megabyte, making 2026 budgets an order of magnitude more generous than the first versions sketched in 2017.

The bad news is that payloads are growing faster. The median mobile page is now 2.6 MiB, having passed the size of DOOM (2.48 MiB) in April. The 75th-percentile page outweighs two copies of DOOM, and pages beyond the 90th percentile are more than 4.5 times larger still. The post's data shows sizes have roughly doubled at each percentile over the past decade, with the median mobile page more than 2.5 times its 2015 weight and P75 pages four times theirs. Mobile JavaScript alone has more than doubled since 2015, now standing at 680 KiB at the median and 1.3 MiB at P75.

The consequence, according to the latest CrUX data cited in the post: fewer than half of origins pass Core Web Vitals for mobile users, more than 40% still perform poorly on desktop, and progress in both cohorts has plateaued.

Test devices: yesterday's mid-ranger is today's P75

The hardware backdrop explains the reference picks. Smartphone replacement cycles are lengthening, with the median device now lasting more than two years, P75 devices approaching three years, and TechInsights estimating a 23.7% annual replacement rate. The Galaxy A24, a mid-2023 release with a 6 nm SoC that launched at $250 against a $353 global average selling price, is the new benchmark; anything built on MediaTek's Helio G99 or Samsung's Exynos 1330, such as the Galaxy A16 4G, Galaxy A07 4G or Redmi Note 13 Pro 4G, will do. Teams serious about the low end should keep testing on Galaxy A51-class hardware or late-model Moto E devices.

For accurate link throttling, the post recommends the $3 Throttly app on Android, arguing that browser DevTools throttling never reproduces real network conditions. On desktop, the prescription is a sub-$250 Windows laptop with a Celeron chip and eMMC storage, such as the HP 14 dq3500nr, to capture the drag from Defender, NTFS and background services.

Why it matters

The post frames the situation as a business and ethical problem, not just a technical one: budgets define who gets a usable web, and the quarter of users below the P75 baseline are disproportionately those who cannot buy better devices or connections. The author argues moderation will not emerge organically while incentives stay misaligned, and calls on browsers, search engines and developer tools to nudge users away from bloated sites. For individual teams, the actionable core is concrete: test on 2023 budget hardware over a 9 Mbps link, keep critical-path JavaScript inside the roughly 0.6 to 1.15 MiB envelope, and consolidate early connections so the budget you have is not spent on handshakes.