Every number on this page was measured, not estimated: we requested eight thumbnail files for each of 9,055 public YouTube videos directly from YouTube's image servers on 7 September 2026. Paste a link above to see which sizes your own video has, then read what the whole sample shows.
Developers, bloggers, and creators repeat a handful of claims about YouTube thumbnails: that maxresdefault.jpg is missing for "many" videos, that a missing size "returns a placeholder", that the biggest thumbnail is 1280x720, that Shorts have no proper vertical image. Most of those claims circulate without a number attached. This page attaches the numbers. On 7 September 2026 we probed 9,055 public videos from two independent samples and recorded, for each of eight thumbnail files, whether it exists, its HTTP status, its pixel dimensions, and its size in bytes. The full methodology and a per-video CSV are below, and you can reuse everything under CC BY 4.0 with a link back. If you only need one video's thumbnail, the YouTube thumbnail grabber above shows every size it has in one click.
maxresdefault.jpg (1280x720). 89.4% of recent uploads, 86.7% of older videos.hqdefault.jpg (480x360), mqdefault.jpg and default.jpg. Not one of 8,664 videos lacked them.maxresdefault.jpg file size; 90% are under 185.4 KB, the largest was 490.3 KB.oar2.jpg, up from 2.6% of 2022 uploads to 36.2% of 2026 uploads.For every one of the 8,664 live videos we requested the eight files below from i.ytimg.com. A file counts as present when the server answered 200 or 206 with a real image; a 404 counts as missing. The three small sizes are universal, sddefault is nearly so, and the two 1280x720 files are where availability drops.
Share of 8,664 public videos for which each thumbnail file exists on i.ytimg.com, 7 Sep 2026.
| File | Pixels (typical) | All 8,664 videos | Recent uploads (5,154) | Wikidata-referenced (3,510) |
|---|---|---|---|---|
default.jpg | 120x90 | 100.0% | 100.0% | 100.0% |
mqdefault.jpg | 320x180 | 100.0% | 100.0% | 100.0% |
hqdefault.jpg | 480x360 | 100.0% | 100.0% | 100.0% |
sddefault.jpg | 640x480 | 95.8% | 97.3% | 93.6% |
hq720.jpg | 1280x720 | 88.2% | 89.2% | 86.6% |
maxresdefault.jpg | 1280x720 | 88.3% | 89.4% | 86.7% |
vi_webp/…/maxresdefault.webp | 1280x720 | 85.7% | 87.8% | 82.7% |
oar2.jpg | original aspect ratio | 29.9% | 25.6% | 36.3% |
Two details matter for anyone writing a fallback chain. First, hq720.jpg is not an extra source of HD images: it never existed for a video that lacked maxresdefault (0 of 8,664), and in 93.6% of videos it was byte-for-byte the same file size as maxresdefault.jpg. Second, the best file available per video was: maxresdefault for 88.3%, sddefault (640x480) as the largest for 7.5%, and hqdefault (480x360) as the largest for 4.2%. That is the whole case for the maxres, then sd, then hq chain the thumbnail URL guide recommends: it covers 100% of videos with at most two fallbacks.
The recent-uploads sample carries each video's publish date, so we can see how the HD thumbnail rate has changed over time. Early videos almost never have one; anything uploaded since 2023 has one more than nine times out of ten. The low rate for the oldest years is not a sampling accident: uploads from 2008 and 2009 in the sample had a maxresdefault 0% of the time, because HD source video and custom thumbnails were rare then and YouTube did not back-fill.
Share of videos with a maxresdefault.jpg, by year of upload (recent-uploads sample; years with fewer than 30 videos omitted: 2008 to 2011 ranged from 0% to 60% on 10 to 15 videos each).
| Upload year | Videos | Have maxresdefault | Vertical (oar2 9:16) |
|---|---|---|---|
| 2012 | 33 | 48.5% | 0.0% |
| 2016 | 35 | 65.7% | 0.0% |
| 2018 | 67 | 59.7% | 0.0% |
| 2020 | 281 | 81.9% | 0.0% |
| 2021 | 274 | 81.8% | 0.0% |
| 2022 | 303 | 80.9% | 2.6% |
| 2023 | 319 | 91.5% | 21.3% |
| 2024 | 520 | 92.5% | 26.0% |
| 2025 | 803 | 92.8% | 28.3% |
| 2026 | 2205 | 94.6% | 36.2% |
A widely repeated claim is that YouTube serves the grey placeholder "at HTTP 200". That is not what the servers did in this study. Every one of the 1,014 missing maxresdefault.jpg requests came back as HTTP 404, and every one of those 404 responses carried the same 1,097-byte body: a valid 120x90 JPEG, the grey placeholder. We re-checked a sample against img.youtube.com and i3.ytimg.com with a desktop browser user agent and got identical 404s. So the practical trap is real but the mechanism is different from the folklore: an <img> tag happily decodes the 404 body and shows a tiny grey box, and new Image() fires onload with a naturalWidth of 120. Code that only checks "did an image load" is fooled; code that checks the status code, or the decoded width, is not. The Python guide and command-line guide both use the status-plus-size check.
These are the bytes YouTube actually serves after re-encoding the creator's upload, which is why even the largest file is far below the 2 MB limit that applies when you upload a custom thumbnail (see thumbnail too large).
| File | Pixels | Median | Mean | 90th percentile | Largest seen |
|---|---|---|---|---|---|
default.jpg | 120x90 | 3.0 KB | 3.1 KB | 4.4 KB | 12.1 KB |
mqdefault.jpg | 320x180 | 9.7 KB | 10.5 KB | 16.9 KB | 53.3 KB |
hqdefault.jpg | 480x360 | 17.3 KB | 19.5 KB | 34.0 KB | 137.4 KB |
sddefault.jpg | 640x480 | 41.0 KB | 44.7 KB | 70.8 KB | 216.6 KB |
maxresdefault.jpg | 1280x720 | 95.3 KB | 109.7 KB | 185.4 KB | 490.3 KB |
The WebP variant is the quiet win for anyone embedding thumbnails on a page: it existed for 85.7% of videos, and across 7,424 JPEG/WebP pairs its median size was 50.9% of the JPEG's. It was smaller in 99.4% of pairs. The catch is coverage: 226 videos had a JPEG maxresdefault but no WebP, and only 4 had WebP without the JPEG, so request WebP first and fall back to JPEG rather than the other way around. The embed guide shows the <picture> pattern. A re-probe on 19 September extends the comparison to sddefault and hqdefault: the saving holds at 640x480, but at 480x360 the WebP is often the larger file for recent uploads.
Almost. Of the 7,650 videos with a maxresdefault.jpg, 7,571 (99.0%) served exactly 1280x720. But 69 videos served a genuine 1920x1080 file at the same URL. All 69 sit in the Wikidata-referenced sample, which skews toward well-known, often older uploads; none of the 4,607 recent uploads with a maxresdefault exceeded 1280x720. We verified the 1920x1080 files by decoding them fully, not just reading headers. The honest statement is therefore: 1280x720 is the size you will get for practically every video today, and the resolution guide is right that there is no 4K thumbnail file, but a small legacy set of videos does carry a 1080p thumbnail at the maxres URL.
The smaller files were uniform: hqdefault.jpg was 480x360 for 99.6% of videos (a handful of very old uploads returned 480x270 or 440x360), sddefault.jpg was 640x480 except for 22 old videos at 640x360, and mqdefault.jpg was 320x180 every single time. The 4:3 sizes (hqdefault, sddefault) are letterboxed 16:9 frames, which is why the size guide tells designers to work at 1280x720 and ignore the 4:3 files.
oar2.jpg is YouTube's "original aspect ratio" thumbnail. It existed for 29.9% of videos overall, but its meaning depends on the video. For landscape videos it is usually a duplicate 1280x720 (or, for some custom uploads, 1920x1080, 1280x544 cinema-crop, or a square 720x720). For Shorts it is the only file that holds the vertical image: 1,031 vertical videos served 1080x1920 and 123 served 720x1280. Their maxresdefault.jpg, meanwhile, was still a 1280x720 landscape frame in 98.1% of cases, with the vertical picture padded into it. If you download a Short's thumbnail from the "1280x720" URL, you get a pillarboxed image; the Shorts thumbnail downloader pulls oar2.jpg for that reason.
The trend is steep. Among recent uploads, the share with a vertical oar2 went from 0.0% (2021) and 2.6% (2022) to 21.3% in 2023, 26.0% in 2024, 28.3% in 2025 and 36.2% for 2026 uploads. Within 2026 the monthly figure climbed to 41.7% in August and 44.4% in the first week of September. On the channels in this sample, roughly two in five new uploads are now vertical.
Share of 2026 uploads (recent-uploads sample, n = 2,205) that have a vertical 9:16 oar2.jpg, by upload month.
Share of uploads with a vertical oar2.jpg by upload year, 2021 to 2026.
Nothing new was probed for this section. It re-reads the same 8,664-row dataset from 7 September to answer two questions readers sent after the study went up: are HD thumbnails getting heavier, and can you treat oar2.jpg as "the Shorts thumbnail"? The short answers are yes for 2026 uploads, and no.
Restricting the recent-uploads sample to landscape videos that have a maxresdefault.jpg (vertical Shorts are excluded here because their pillarboxed frame compresses differently, see below), the median file was between 94.9 KB and 113.5 KB for every upload year from 2018 to 2025, with the 2020 to 2025 pool at 101.5 KB (n = 1,795). For 2026 uploads the median is 136.9 KB (n = 1,293), the 90th percentile is 259.5 KB against 175.7 KB for 2020 to 2025, and 43.8% of files exceed 150 KB against 17.3%. Month by month the step is visible from February 2026 onward (143.7 KB in February, 119.5 to 126.0 KB in April to June, 144.4 to 152.4 KB from July to the first week of September; every month n ≥ 58). The dataset cannot say why: it could be a change in how YouTube encodes new thumbnails, or newer files could carry a higher-quality encode that gets recompressed later. Either way, if you embed thumbnails of new videos, budget roughly 140 to 150 KB per HD JPEG rather than the 95 KB overall median, which makes the WebP variant (about half the bytes) worth more, not less.
Median maxresdefault.jpg size in KB for landscape videos, by upload year (recent-uploads sample, years with at least 40 such videos; 2013 had 45 at 87.3 KB and is omitted for the gap).
Every one of the 1,222 vertical videos with a maxresdefault.jpg served it at 1280x720, the vertical picture padded with black on both sides. Black bars cost almost nothing in JPEG, so the median vertical maxresdefault was 83.8 KB against 98.3 KB for landscape videos across the whole dataset, and among 2026 uploads 87.5 KB (n = 792) against 136.9 KB (n = 1,293). The same holds for hqdefault.jpg: 11.5 KB vertical against 18.9 KB landscape. Per video, sddefault.jpg was on median 42.9% of the maxresdefault bytes and hqdefault.jpg 18.4% (7,650 videos with both files). Across all 7,650 maxresdefault files, 8.7% were under 50 KB, 45.1% between 50 and 100 KB, 27.0% between 100 and 150 KB, 11.2% between 150 and 200 KB, and 7.9% above 200 KB.
The original study reported that 29.9% of videos have an oar2.jpg. Split by sample, the file means two different things. Among the 1,321 recent uploads that have one, 93.7% are vertical (1,238 videos). Among the 1,273 Wikidata-referenced videos that have one, 0.5% are vertical (7 videos); the rest are landscape copies at 1280x720 (825), 640x480 (79), a square 720x720 (70), 1920x1080 (49), a 1280x544 cinema crop (28) or 854x480 (26). So a script that treats "oar2 exists" as "this is a Short" is right about 94% of the time for new videos and wrong almost every time for older, well-known ones. Read the image's width and height and call it vertical only when height exceeds width, which is what the Shorts thumbnail downloader and the browser extension do. By upload year in the recent sample, the landscape-oar2 share has stayed between 0.4% and 4.4% since 2020 while the vertical share climbed from 0% to 36.2%.
Share of uploads whose oar2.jpg is vertical (dark, first number) or landscape (light, second number), by upload year, recent-uploads sample 2020 to 2026.
The original study compared WebP with JPEG at one size only, maxresdefault, and found WebP smaller in 99.4% of pairs. That left open the question tool authors actually argue about (yt-dlp's issue tracker has a long thread titled "jpg thumbnails are actually better than webp?"): does the saving hold at the smaller sizes most embeds and players use? On 19 September 2026 we drew 1,000 videos from the dataset that have a maxresdefault.jpg (500 from the recent-uploads sample, 500 from the Wikidata sample, seeded random draw) and sent a HEAD request for each video's maxresdefault.webp, sddefault.webp and hqdefault.webp under https://i.ytimg.com/vi_webp/VIDEO_ID/, recording the status and Content-Length. The JPEG bytes are the 7 September measurements of the same videos. 3,000 requests, zero errors. The per-video rows are in youtube-thumbnail-webp-vs-jpeg-2026.csv.
| Size | WebP exists | WebP missing (JPEG exists) | Median JPEG | Median WebP | Median WebP ÷ JPEG | WebP larger than JPEG |
|---|---|---|---|---|---|---|
maxresdefault (1280x720) | 965 | 35 (3.5%) | 96.7 KB | 50.8 KB | 0.53 | 10 (1.0%) |
sddefault (640x480) | 970 | 30 (3.0%) | 41.3 KB | 23.3 KB | 0.55 | 14 (1.4%) |
hqdefault (480x360) | 970 | 30 (3.0%) | 19.0 KB | 15.6 KB | 0.75 | 324 (33.4%) |
1,000 videos with a maxresdefault.jpg; WebP measured 19 September 2026, JPEG 7 September 2026. Medians are taken over the videos where both files exist; the ratio column is the median of the per-video ratios.
At the two larger sizes the saving is consistent: summed over all 965 pairs, the WebP maxresdefault files total 59.1% of the JPEG bytes (the 90th percentile drops from 192.8 KB to 132.1 KB), and at sddefault 59.5%. At hqdefault the picture splits by age. For the Wikidata-referenced videos, which skew older, the median ratio is 0.65 and WebP is the larger file for 10.4%. For the recent uploads the median ratio is 1.06 and WebP is larger for 56.1%; by upload year that share climbs from 25% (2020, median ratio 0.69) through 32% (2022, 0.79) and 43% (2024, 0.89) to 55% (2025, 1.05) and 72% for 2026 uploads (n = 217, median ratio 1.13). Vertical Shorts are the extreme: their pillarboxed hqdefault WebP is larger than the JPEG in 85.3% of cases (n = 116, median ratio 1.17), while their maxresdefault WebP is only 0.40 of the JPEG. Summed over every hqdefault pair the WebP files still total 82.7% of the JPEG bytes, because the older videos' savings are large and the recent videos' penalties are small (median JPEG 16.7 KB against WebP 16.5 KB in the recent sample).
Two things this does not say. Bytes are not quality: the two files are separate encodes of the same source frame, YouTube does not publish its encoder settings, and this dataset has no fidelity measure, so a heavier WebP may simply be a higher-quality encode. And it does not say why the hqdefault ratio has drifted upward for newer uploads; the data shows the trend, not the cause. What it does settle is the byte question: if you choose a format to save bandwidth, WebP wins at 1280x720 and 640x480 for practically every video, and at 480x360 it is more often the heavier file for videos uploaded since 2025.
30 of the 1,000 videos had no WebP file at any of the three sizes, and 5 more had sddefault.webp and hqdefault.webp but no maxresdefault.webp. Every miss was HTTP 404 carrying a 552-byte image/webp body, the WebP twin of the 1,097-byte grey JPEG. Recent uploads lacked the WebP maxresdefault less often (12 of 500) than Wikidata-referenced videos (23 of 500). The fallback advice is the same as for JPEG: check the status code or the decoded width rather than the load event, and fall back from .webp to .jpg.
Limitations of the re-probe. The sample is restricted to videos that have a JPEG maxresdefault, so it says nothing about WebP for the 11.7% of videos that lack one. JPEG and WebP sizes were measured twelve days apart; a video whose thumbnail was replaced in between would compare two different frames. The WebP files' pixel dimensions were not read, only status and size.
Two independent samples of public video IDs, both from open sources, no YouTube Data API key and no scraping of YouTube pages:
default.jpg at all, meaning the video has since been deleted or made private, and were excluded from every percentage.youtube.com/feeds/videos.xml?channel_id=; the YouTube RSS feed generator writes that URL for any channel) of 394 channels whose channel ID is stored in Wikidata property P2397, up to 15 videos per channel. Publish dates range from 2008-03-12 to 2026-09-07; 2,205 videos were uploaded in 2026. This sample skews toward established channels (they have a Wikidata entry) and toward whatever those channels posted most recently, which is why it is Shorts-heavy.For each video we requested eight URLs from https://i.ytimg.com/vi/VIDEO_ID/: default.jpg, mqdefault.jpg, hqdefault.jpg, sddefault.jpg, hq720.jpg, maxresdefault.jpg, oar2.jpg, plus https://i.ytimg.com/vi_webp/VIDEO_ID/maxresdefault.webp. Each request asked for the first 4,096 bytes (an HTTP Range request), which is enough to read the JPEG frame header for pixel dimensions while the Content-Range header reports the full file size. Status 200 or 206 counts as present; 404 counts as missing. The 1920x1080 maxresdefault files were downloaded in full and decoded to confirm the header reading. All 9,055 videos and 72,440 requests completed with zero network errors between roughly 16:00 and 18:30 UTC on 7 September 2026. WebP dimensions were not parsed, only presence and size.
Limitations. Neither sample is a uniform random draw from all of YouTube; that is not possible without the API and would include vast numbers of near-zero-view uploads. Both samples over-represent notable channels, so the maxresdefault rate for the long tail of casual uploads is probably lower than reported here. Percentages are for the day of measurement; YouTube generates thumbnails asynchronously and a video uploaded hours before probing may gain files later. "Vertical" is defined as oar2.jpg having a height greater than its width, which captures Shorts and any other portrait upload.
Data and charts on this page are released under CC BY 4.0. Suggested citation: Thumbnails Grabber (2026). YouTube Thumbnail Statistics 2026: Data From 8,664 Videos. https://thumbnailsgrabber.com/youtube-thumbnail-statistics. Data collected 7 September 2026. Per-video dataset: youtube-thumbnail-availability-2026.csv (8,664 rows: video ID, cohort, publish date, presence flag for each of the eight files, maxresdefault status, width, height and bytes, sddefault and hqdefault bytes, oar2 width and height). Re-probe dataset (19 September 2026): youtube-thumbnail-webp-vs-jpeg-2026.csv (1,000 rows: video ID, cohort, and for maxresdefault, sddefault and hqdefault the WebP HTTP status, WebP bytes and JPEG bytes).
A developer-focused write-up of these findings is published on dev.to, and the same fallback logic answers a Stack Overflow question about picking the highest-quality YouTube thumbnail in React. The site's editorial rules are on the About page.
maxresdefault.jpg, then sddefault.jpg, then hqdefault.jpg; that chain resolved to a real image for 100% of videos.naturalWidth > 120 (browser side).oar2.jpg, and around 40% of new uploads in this sample now need it.In this study of 8,664 public videos measured on 7 September 2026, maxresdefault.jpg existed for 88.3% overall: 89.4% of recent uploads and 86.7% of older, Wikidata-referenced videos. For uploads from 2023 onward the rate is 91 to 95 percent; for videos uploaded in 2008 and 2009 it was 0 percent.
An HTTP 404 status, every time: all 1,014 missing maxresdefault requests in the study returned 404. The body of that 404 is still a 1,097-byte, 120x90 grey JPEG, so an <img> tag renders a tiny placeholder instead of showing a broken image. Code should check the status code or the decoded width, not just whether an image appeared.
default.jpg (120x90), mqdefault.jpg (320x180), and hqdefault.jpg (480x360) existed for 100% of the 8,664 live videos. sddefault.jpg (640x480) existed for 95.8%. The safe fallback chain is maxresdefault, then sddefault, then hqdefault.
The median maxresdefault.jpg was 95.3 KB, the 90th percentile 185.4 KB, and the largest seen 490.3 KB. hqdefault.jpg had a median of 17.3 KB and default.jpg about 3.0 KB. These are the files YouTube serves, which are re-encoded from the creator's upload, so they are far under the 2 MB upload limit.
At 1280x720, yes: the WebP maxresdefault at /vi_webp/VIDEO_ID/maxresdefault.webp existed for 85.7% of videos and its median file size was 50.9% of the matching JPEG, smaller in 99.4% of the 7,424 pairs compared. A 1,000-video re-probe on 19 September 2026 found the same at 640x480 (median 0.55 of the JPEG bytes) but not at 480x360, where the median was 0.75 and the WebP hqdefault was the larger file for 33% of videos overall and 72% of 2026 uploads.
Yes, at oar2.jpg. Among the 5,154 recent uploads, 24.0% had a vertical oar2.jpg, usually 1080x1920 (1,028 videos) or 720x1280 (121). Their maxresdefault.jpg was still a 1280x720 landscape file in 98.1% of cases, so a landscape download of a Short shows the vertical image with padding; the true vertical image is oar2.jpg.
For 7,571 of the 7,650 videos with a maxresdefault (99.0%) the file was 1280x720. 69 videos, all in the older Wikidata-referenced cohort, served a 1920x1080 maxresdefault.jpg, and none of the recent uploads did. So 1280x720 is the practical ceiling, with a small number of legacy exceptions.
Yes. The aggregate figures and the per-video CSV are published under CC BY 4.0. Cite it as: Thumbnails Grabber, "YouTube Thumbnail Statistics 2026", https://thumbnailsgrabber.com/youtube-thumbnail-statistics, data collected 2026-09-07.
Want to see the image first? Paste any link into the YouTube thumbnail grabber and download every size in one click.