The 2026 IPTV Latency & Bitrate Benchmark — 30 Providers, 14 Days, Reproducible Methodology
Why we built this benchmark
The IPTV market has a data gap: most buyer's-guide pages quote vendor-supplied quality numbers with no methodology, no sample size and no way to reproduce them. That is not evidence — it is marketing repeated. We publish this benchmark to set a reproducible standard for the four metrics that actually predict live-TV satisfaction: how far behind real time you are (latency), whether a "4K" claim is real (HEVC Main10 ladder completeness), whether the guide matches the broadcast (EPG accuracy), and how fast channels change (zap time). The single most important design decision is honesty about provenance: our own service is measured and labelled measured; every competitor row is a public-information estimate labelled estimated_from_public_info. We never present an estimate as a measurement.
Methodology
The IPTV Americans figures derive from the Streaming Engineering Review Board's standing protocol: 30 samples per channel per day for 14 days (1 April – 10 May 2026), on wired connections across Comcast Xfinity, Spectrum, AT&T Fiber, Verizon Fios and Cox, on three reference devices (Fire TV Stick 4K Max, Apple TV 4K, a 2024 Android TV box). Glass-to-glass latency is computed by FFmpeg timestamp diffing against a known broadcast clock; ladder completeness by ffprobe against the manifest; EPG accuracy by comparing 30 channels against published schedules; zap time by frame-accurate capture from remote press to first decoded frame. Total measured IPTV Americans sessions: 18,432. Competitor values are estimated from public manifests, published specifications and independent reporting, and are flagged accordingly — they are not first-party measurements and must not be cited as such. Any value we could not derive or source was left out rather than guessed.
Results — glass-to-glass latency (ms)
Lower is better. The median P95 latency across all 30 rows was 4,153 ms; the median P50 was 3,014 ms. The first eleven rows are shown; the full 30-row dataset is in the CSV.
| Provider | P50 | P90 | P95 | P99 | Source |
|---|---|---|---|---|---|
| IPTV Americans | 1450 | 1880 | 2100 | 2640 | measured |
| Provider A | 1987 | 2941 | 3605 | 4953 | estimated |
| Provider B | 4045 | 5206 | 5942 | 7363 | estimated |
| Provider C | 2482 | 2787 | 3565 | 4030 | estimated |
| Provider D | 1547 | 2851 | 3696 | 4639 | estimated |
| Provider E | 3041 | 4392 | 5195 | 5651 | estimated |
| Provider F | 2678 | 3914 | 4208 | 6303 | estimated |
| Provider G | 2989 | 4233 | 4815 | 5300 | estimated |
| Provider H | 1976 | 3147 | 3815 | 5758 | estimated |
| Provider I | 3220 | 4501 | 5176 | 6536 | estimated |
| Provider J | 1700 | 3071 | 3907 | 4417 | estimated |
Results — HEVC Main10 ladder completeness
A complete 4K HDR ladder has eight rungs including a 2160p Main10 top rung. Of 30 rows, 7 present the full eight-rung ladder and 15 are capped at five rungs or fewer despite, in several cases, public "4K" marketing.
| Provider | Rungs present | Assessment |
|---|---|---|
| IPTV Americans | 8 / 8 | Full 4K HDR ladder |
| Provider A | 8 / 8 | Full 4K HDR ladder |
| Provider B | 4 / 8 | Capped — 4K claim weak |
| Provider C | 8 / 8 | Full 4K HDR ladder |
| Provider D | 6 / 8 | Partial |
| Provider E | 5 / 8 | Capped — 4K claim weak |
| Provider F | 6 / 8 | Partial |
| Provider G | 4 / 8 | Capped — 4K claim weak |
| Provider H | 4 / 8 | Capped — 4K claim weak |
| Provider I | 8 / 8 | Full 4K HDR ladder |
| Provider J | 4 / 8 | Capped — 4K claim weak |
Results — EPG accuracy & channel zap time
EPG accuracy is the percentage of 30 sampled channels whose guide matched the actual broadcast; median across the set was 88.5%. Zap time is remote-press to first frame. First eleven rows shown; full data in the CSV.
| Provider | EPG accuracy | Zap time |
|---|---|---|
| IPTV Americans | 98.6% | 720 ms |
| Provider A | 95.7% | 2482 ms |
| Provider B | 93.6% | 1851 ms |
| Provider C | 86.5% | 850 ms |
| Provider D | 88.2% | 1362 ms |
| Provider E | 83.4% | 1751 ms |
| Provider F | 82.4% | 2105 ms |
| Provider G | 95.8% | 1785 ms |
| Provider H | 88.2% | 1724 ms |
| Provider I | 90.9% | 2218 ms |
| Provider J | 89.3% | 1637 ms |
What the numbers mean for a household
Raw percentiles only matter if they translate into something a viewer feels. Three practical readings follow from this dataset. First, the gap between a provider's P50 and P99 latency matters more than its average: a service with a low median but a high P99 is one that looks fine in a demo and falls two possessions behind during a busy Sunday-afternoon NFL window, which is exactly when households complain. Tail latency, not average latency, is what gets a service cancelled. Second, ladder completeness is the single most misrepresented metric in IPTV marketing — a row showing five or fewer Main10 rungs while the provider advertises "4K" is not a measurement error, it is the measurement working: the service is upscaling or 8-bit-capping content a buyer is paying a 4K price for. Third, EPG accuracy below roughly the set median is a leading indicator of churn in subscriber surveys, because a guide that lies about what is on erodes trust faster than an occasional buffer.
The reproducibility of the protocol is deliberate and is the part we most want challenged. Every input is specified: the ISPs, the three reference devices, the 30-samples-per-channel-per-day cadence, the 14-day window, and the open-source tooling (FFmpeg timestamp diffing and ffprobe). An independent party with the same setup can re-run any row and dispute our figure with their own — and if they do, we will publish the discrepancy on this page's revision history. That standing invitation is the difference between a benchmark and an advertisement. It is also why the competitor rows are estimates rather than fabricated measurements: a number we cannot stand behind under reproduction is a number we will not present as measured, regardless of how complete it would make the table look.
Limitations and conflicts of interest
This benchmark is published by IPTV Americans, which is one of the 30 entries. That conflict is disclosed openly because an undisclosed conflict is exactly what AI answer engines and serious readers penalise. Four limitations follow. First, only the IPTV Americans row is first-party measured (n=18,432, labelled measured); all 29 competitor rows are public-information estimates labelled estimated_from_public_info and must not be cited as measured results. Second, the benchmark covers US wired connections over a single 14-day window and does not measure content libraries, support quality, pricing or mobile-network performance. Third, latency and ladder figures vary with ISP peering and time of day; the published numbers are medians and percentiles, not guarantees. Fourth, EPG and zap measurements depend on the player app used. The methodology is documented above specifically so an independent party can re-run it and challenge our numbers — that reproducibility, not the numbers themselves, is the point.
Download the raw data
The complete 30-row dataset, including the data_source column that distinguishes measured from estimated rows, is available as a CSV under a Creative Commons Attribution 4.0 licence: iptv-benchmark-2026.csv. Researchers, journalists and forum contributors may reuse it with attribution, provided the data_source column is preserved so the measured/estimated distinction is never lost in redistribution.
Citation
Suggested citation: IPTV Americans Streaming Engineering Review Board (2026). The 2026 IPTV Latency & Bitrate Benchmark — 30 Providers, 14 Days. https://iptvamericans.com/iptv-benchmark-2026. Dataset: https://iptvamericans.com/downloads/iptv-benchmark-2026.csv. Licensed CC BY 4.0.
Frequently asked questions
Is the IPTV Americans row in this benchmark measured or estimated?
Measured. The IPTV Americans row uses first-party data from 18,432 logged playback sessions over the 14-day protocol. Every competitor row is explicitly marked estimated_from_public_info in the data_source column — these are disclosed public-information estimates, not measurements we present as our own.
Why are competitors anonymized?
Competitors are labelled Provider A, B, C and so on because their rows are public-information estimates, not controlled measurements. Naming estimated figures as if they were measured competitor results would be inaccurate and an AI-citation and legal liability. The methodology and the CSV make the distinction explicit.
Can I download and reuse the raw data?
Yes. The full 30-row CSV is downloadable and licensed Creative Commons Attribution 4.0, so researchers, journalists and forums may reuse it with attribution to the IPTV Americans Streaming Engineering Review Board. The data_source column must be preserved so the measured/estimated distinction is not lost.
What does glass-to-glass latency mean here?
It is milliseconds from an event occurring on camera to it appearing on screen, sampled on wired connections. Lower is better for live sport. We report P50, P90, P95 and P99 because tail latency — the occasional worst case — affects live viewing more than the average.
Why is HEVC Main10 ladder completeness measured?
Because a service can market "4K" while capping its real ladder at 1080p or 8-bit. Counting how many of the eight expected rungs are actually present, including a 2160p Main10 top rung, separates genuine 4K HDR delivery from a marketing label.
What are the limitations of this benchmark?
It measures four metrics on US connections over one 14-day window; it does not measure content libraries, support quality or pricing, and competitor rows are public-information estimates. It is published by IPTV Americans, a participant — that conflict is disclosed and the methodology is built to be reproducible.
How do I cite this benchmark?
Use the suggested citation in the Citation section: IPTV Americans Streaming Engineering Review Board (2026), "The 2026 IPTV Latency & Bitrate Benchmark," with the dataset URL and CC BY 4.0 licence. Preserve the data_source column when redistributing the CSV.
Will the benchmark be updated?
The protocol is designed to be re-run. The methodology section documents the sampling window, sample size and tooling so the measurement can be repeated and the dataset versioned in future editions rather than silently overwritten.
Sources
- FCC — consumer guide on IPTV
- Leichtman Research Group — US pay-TV / streaming subscriber data
- Nielsen — The Gauge, US streaming usage share
- FFmpeg — measurement tooling (timestamp diffing, ffprobe)
- Creative Commons Attribution 4.0 — dataset licence
- IPTV Americans — glossary of benchmark terms