Ranked on Merit

Scoring sheet

How we scored Plex, Jellyfin and Emby

Every score in Plex vs Jellyfin vs Emby comes from this sheet: the tools we considered, the scale for each criterion, and the evidence and date behind each score.

This page shows the working behind Plex vs Jellyfin vs Emby: the criteria and weights, what each score means, and for every server and criterion the score and the evidence behind it.

We ran all three servers ourselves on October 28, 2026: Plex Media Server 1.43.5, Jellyfin 12.4.0 and Emby Server 4.10.2, in their official Docker images on a Linux cloud machine with 4 virtual CPUs, 15 GB of memory and no graphics card. Setup, metadata matching and software transcoding are our own test results. Apps, prices, remote access, hardware transcoding support and privacy terms come from each company's own pages, read the same day. We did not use the TV and phone apps, test hardware transcoding or test remote access over a real home connection.

The question and the reader

The question, in the reader's words: "Plex vs Jellyfin vs Emby: which media server should I use?" We asked four AI models (GPT, Claude, Gemini and Perplexity, through their APIs with web search on) five versions of it in October 2026: the head-to-head, whether Plex is still worth it, the best free media server, whether you need Plex Pass or Emby Premiere to watch away from home, and which is easiest for non-technical family members. That gave 20 answers.

The reader has films and TV shows as files on a computer or a network drive and wants to watch them on the TVs, phones and tablets in the house, and often away from home or at relatives' homes. Many are Plex users wondering whether to leave after its 2025 and 2026 price changes. They want to know what each one costs, how hard it is to set up and share, and whether it will name and organize their files properly.

The field

The question names three servers, so these three are scored. The AI answers also named Kodi, Infuse, Channels DVR, Plexamp, Serviio and Jellyfin apps such as Swiftfin and Streamyfin; Stremio appeared in the title of a page one answer cited, and we added Navidrome and Universal Media Server from web searches for alternatives. None of those is a general media server that streams your own video to other screens: Kodi and Infuse are players that can use these servers as their library, Plexamp and Navidrome are for music, Channels DVR is built around recording live TV, Stremio streams add-ons, and Serviio and Universal Media Server are DLNA servers without remote apps or accounts. The page lists each with its reason.

Rubric

We set the criteria, weights and anchors at about 22:20 UTC on 28 October 2026, and the rubric is recorded at that time in our version history. As written then: "At that point we had read the three vendors' pricing and feature pages and their app lists, run each server's first-run setup, and seen that Plex Media Server could not fetch any metadata before it was signed in to a Plex account. No metadata-matching or transcoding test had run."

Note added later the same evening: the belief about Plex was wrong. Plex had failed on a probe file because the test machine's network intercepts secure connections with its own certificate, which Plex did not trust; once Plex was given that certificate, it fetched metadata for every file without signing in. Also, only Jellyfin's and Emby's setup wizards had been run; Plex's web app goes straight to a sign-in page. Neither point changed a criterion, weight or anchor. Both are listed under "What changed after the rubric was written".

Criterion Weight Why this weight
Apps and devices 25 The reader wants their films on every TV in the house. A server is only as useful as the apps on the screens they own.
Cost 25 "Is Plex still worth it?" is a price question, and the three range from free to a $749.99 lifetime pass.
Watching away from home 15 Most households want to watch on a trip or share with family elsewhere, and this is where the three differ most in effort and safety.
Library and metadata 15 Posters, titles and episode names are what make a pile of files feel like Netflix. Wrong matches are the most common complaint.
Transcoding 10 Matters when a TV or phone can't play a file as it is, or a remote connection is slow. Many households rarely need it.
Setup and privacy 10 Whether you need an online account, and what the server shares, matters to many readers but rarely decides on its own.

Apps and devices (25)

Twelve kinds of screen a household might watch on: a web browser, a Windows or Mac computer, an iPhone or iPad, an Android phone, Android TV or Google TV, Amazon Fire TV, Roku, Apple TV, Samsung TV, LG TV, Xbox and PlayStation.

For each: 1 point if there is an official app that plays your own media at home with no payment; 0.5 if an official app exists but full playback at home needs a payment (an app unlock or a subscription); 0 if there is no official app. Score = points / 12 x 10, to one decimal. Third-party apps are named on the page but not counted.

  • 10: an official, free app on all 12
  • 7: about 8.5 points
  • 4: about 5 points
  • 0: none

Cost (25)

C = the lowest US price, over five years, to unlock everything this reader is most likely to pay for: remote streaming for every member of the household, hardware-accelerated transcoding, and downloads to phones. Prices from each vendor's US pricing page on the day of the check.

score = 10 - C / 30, to one decimal, floored at 0.

  • 10: free
  • 7: $90 over five years
  • 4: $180
  • 0: $300 or more

Watching away from home (15)

A checklist, 10 points. What it takes to watch outside the home once any payment is made (price is scored under Cost):

  • 3: remote playback works through a relay when the router's ports can't be opened, with no extra software on the viewer's side.

  • 2: automatic router setup (UPnP or NAT-PMP) is built in.

  • 3: remote connections are encrypted (HTTPS) out of the box, without buying a domain or installing a certificate.

  • 1: remote apps find the server through a sign-in, without typing an address.

  • 1: a TV app can be signed in with a short code instead of typing a password.

  • 10: all five

  • 7: no relay, everything else

  • 4: automatic router setup and one convenience

  • 0: none; remote access is entirely up to you

Library and metadata (15)

Our test library: 47 movie files and 13 TV episode files, named the way people's files are really named (clean "Title (Year)" names, release-style names, missing years, remakes and same-title films, accents, numbers, two versions of one film, a file holding two episodes, an anime with absolute numbering). The right answer for each file is its IMDb ID, checked against Wikidata before the test.

A movie file counts as right when the server matched it to the right film. An episode file counts as right when the server matched the right series and the right season and episode number, and the episode has its real title (not one made from the file name).

score = 10 x (files right / 60), to one decimal.

  • 10: all 60 right
  • 7: 42 right
  • 4: 24 right
  • 0: none

The rubric as first written said 61 files (14 episodes); the manifest, fixed before the test, has 60 (13 episodes). We corrected the count before adding any library; the formula did not change.

Transcoding (10)

Two parts:

  • Up to 7 points for measured software transcoding speed on our 4-core cloud machine (no GPU), with each server's default transcoding settings and throttling turned off so it runs flat out. Three test files: a 4K HEVC 10-bit file to 1080p H.264, a 1080p H.264 file to 720p with English subtitles burned in, and a 1080p HEVC file to 1080p H.264. Speed = seconds of video produced per second of wall time, median of three runs. For each file, a server's ratio is its speed divided by the fastest server's speed; points = 7 x the geometric mean of its three ratios.

  • Up to 3 points for hardware transcoding support in the vendor's documentation (whatever it costs, since price is scored under Cost): 1 each for Intel Quick Sync, NVIDIA and AMD GPUs; 0.5 where the vendor supports a GPU family only "as is" or untested.

  • 10: the fastest on every file, and documented support for all three GPU families

  • 7: about 70% of the fastest speed, and all three GPU families

  • 4: about half the fastest speed, and two GPU families

  • 0: no transcoding

Setup and privacy (10)

A checklist, 10 points:

  • 3: the server can be set up and used without creating an online account (tested).

  • 1: it fetches posters, titles and descriptions without any account (tested).

  • 2: its source code is open.

  • 1: the home screen shows only your own media by default, with no ads or free streaming rows.

  • 2: the vendor's privacy policy or documentation says the server sends no usage data to the vendor by default.

  • 1: playback at home keeps working when the internet connection or the vendor's sign-in service is down.

  • 10: all six

  • 7: no open source code and one other point missing

  • 4: needs an online account but sends nothing else

  • 0: none

Scores

Jellyfin: total 8.4

Criterion Score Evidence
Apps and devices 9.2 11 of 12. Official apps listed on Jellyfin's client page, all free: web app built into the server, Jellyfin Media Player for computers, Jellyfin for iOS, Jellyfin for Android, Jellyfin for Android TV ("for Android TV and Fire TV devices"), Jellyfin for Roku, Swiftfin for Apple TV (labeled "Official Beta"), Jellyfin for Tizen (Samsung), Jellyfin for WebOS (LG) and Jellyfin for Xbox (1 each). No PlayStation app (0). 11 / 12 x 10 = 9.17
Cost 10 Free. There is no paid tier; hardware transcoding, downloads and live TV are part of the free server (Jellyfin's hardware acceleration docs). C = $0
Watching away from home 1 No relay (0). No automatic router setup (0): in the 12.4.0 server we ran, the dashboard has no port-mapping option and the server contains no port-mapping library; its network settings file still has an unused EnableUPnP entry. No HTTPS out of the box (0): Jellyfin's networking docs "strongly recommend" handling HTTPS on a reverse proxy and call opening a port directly "insecure and not recommended". No central sign-in that finds the server (0); automatic discovery "only works locally". Quick Connect signs a TV in with a code approved from a signed-in device (1)
Library and metadata 9.8 59 of 60 right with the web interface's default library settings. Miss: Blade Runner 2049 (2017).mkv matched Blade Runner (1982). 59 / 60 x 10 = 9.83
Transcoding 9.8 6.83 of 7 for speed (see the table below), plus 3 of 3 for hardware: Intel (QSV and VA-API), NVIDIA (NVENC) and AMD (AMF and VA-API) are all documented (hardware acceleration docs)
Setup and privacy 10 Set up and used with no online account: the six-screen wizard asks for a local admin user only (tested, 3). Fetched posters, titles and episode names with no account (tested, 1). Open source (2). Home screen shows only your libraries (tested in the web app, 1). Jellyfin's site says it has no tracking and no central servers collecting data (jellyfin.org, 2). Accounts live on your server, and sign-in and playback worked with the server cut off from the internet (tested on a Docker network with no route out, 1)

Emby: total 7.2

Criterion Score Evidence
Apps and devices 7.5 9 of 12, from Emby's Premiere feature matrix. Free full playback: web browser; Android TV and Fire TV ("Playback for up to 5 TV devices"); Roku; Apple TV; Samsung and LG smart TVs (1 each, 6). Full playback needs payment: the Windows and macOS apps ("Limited Playback (one-minute only)" free, full with Premiere); iPhone and iPad and Android phones (full with an App Unlock or Premiere); Xbox and PlayStation 4 (one minute free, full with Premiere) (0.5 each, 3). 9 / 12 x 10 = 7.5
Cost 6 C = $119: Emby Premiere lifetime, which covers the household (up to 30 devices) and unlocks every app, hardware transcoding and downloads (Emby Premiere, feature matrix). $54 a year would be $270 over five years; $4.99 a month, $299.40. Remote streaming itself needs no payment. 10 - 119 / 30 = 6.03
Watching away from home 4 No relay: "all connections are direct from your devices to your Emby Server" (Emby Connect, 0). Automatic router setup, ticked by default in the setup wizard (tested, 2). No HTTPS out of the box; it needs your own certificate (0). Emby Connect, free, helps apps "locate your Emby Server" (1). TV apps sign in with a PIN entered at emby.media/pin, which uses Emby Connect (Emby forum, 1)
Library and metadata 9.3 56 of 60 right with default library settings. Misses: WALL-E (2008).mkv matched "WALL·E's Treasures & Trinkets", a 2021 short; Star.Wars.1977.mkv matched "The Making of Star Wars"; Battlestar Galactica (2003) S01E01 matched the right series but got no episode metadata; One Piece episode 001 was filed as special No. 1 (an early OVA), not episode 1. 56 / 60 x 10 = 9.33
Transcoding 9.8 6.85 of 7 for speed, plus 3 of 3 for hardware: Intel Quick Sync, NVIDIA NVENC, and AMD through AMF on Windows and VA-API on Linux (Emby's hardware acceleration overview, Windows, Linux). Hardware transcoding needs Premiere except on NVIDIA Shield and WD NAS devices
Setup and privacy 8 Set up and used with no online account: the six-screen wizard asks for a local admin user; Emby Connect is optional (tested, 3). Fetched metadata with no account (tested, 1). Closed source since 2018 (0). Home screen shows your libraries, with a "Get Emby Premiere" button (tested, 1). Emby removed telemetry and anonymous usage reporting in Emby Server 4.0 (Emby's announcement, 2). Accounts live on your server, and sign-in and playback worked with the server cut off from the internet (tested on a Docker network with no route out, 1)

Plex: total 6.9

Criterion Score Evidence
Apps and devices 10 12 of 12. Plex's apps and devices page lists the web app, Windows and macOS, iOS, Android, Google TV and Android TV, Amazon Fire TV, Roku, Apple TV, Samsung and LG among its smart TVs, Xbox One and Series X/S, and PlayStation 4 and 5. Plex's plans page says the apps stream on your home network with "no subscription required" (1 each)
Cost 1.7 C = $249.99: the five-year Plex Pass, which covers remote streaming for everyone using your server, hardware transcoding and downloads (plans page, read October 28, 2026). The alternatives cost more over five years: $69.99 a year ($349.95), $6.99 a month ($419.40) or $749.99 lifetime. A Remote Watch Pass ($29.99 a year per viewer) covers remote streaming only, not hardware transcoding or downloads (Plex free vs paid). 10 - 249.99 / 30 = 1.67
Watching away from home 10 Relay when a direct connection fails, capped at 2 Mbps per stream for both Plex Pass and Remote Watch Pass holders (Plex's relay article) (3). Automatic router setup: "The server will attempt to automatically configure a connection through your router using UPnP or NAT-PMP" (remote access guide, 2). Encrypted remote connections through a certificate Plex issues to each signed-in server under plex.direct (Filippo Valsorda's account of Plex's design, 3). Apps find servers shared with your Plex account (1). TVs sign in with a code at plex.tv/link (1). Remote access needs the server signed in to a Plex account
Library and metadata 10 60 of 60 right with default agents, on a server that was never signed in; we added the libraries through the server's programming interface on the machine itself, because its web app asks for a sign-in first. The two-episode Friends file is listed under both episodes
Transcoding 9 6.52 of 7 for speed, plus 2.5 of 3 for hardware: Intel Quick Sync and NVIDIA supported, AMD supported only "as is": "Support for AMD GPUs is provided 'as is' and your mileage may vary" (Plex's hardware-accelerated streaming article) (1 + 1 + 0.5). Hardware transcoding needs Plex Pass (free vs paid)
Setup and privacy 1 Not usable without an online account: the server's web app went straight to a Plex sign-in page with no way to skip (tested, 0). Fetched metadata for all 60 files while unclaimed (tested, 1). Closed source (0). Apps show Plex's free movies, shows and live TV with your libraries until you turn off Online Media Sources (free live TV guide, 0). The privacy policy (revised May 6, 2026) lists data collected from your server, including configuration data and library data, with optional playback data you can opt out of (0). Plex says "most things will still work fine if you temporarily don't have a connection", but its Apple TV, Android TV, Fire TV, most smart TV, PlayStation and Xbox apps "require an internet connection to use" (internet and network requirements, 0). Plex runs on the host's network in our setup, so it was not part of the offline test

Transcoding results

We asked each server, with its default settings and no graphics card, for three conversions to H.264 video and stereo AAC audio over HLS, with direct play and stream copying turned off:

  • a: Big Buck Bunny re-encoded as 4K HEVC 10-bit (180 seconds), to 1080p at 8 Mbps.
  • b: Sintel, the original 1080p H.264 release (888 seconds, 10 subtitle tracks), to 720p at 4 Mbps with the English subtitles burned in.
  • c: Tears of Steel re-encoded as 1080p HEVC (180 seconds), to 1080p at 8 Mbps.

Our client downloaded the stream's segments in order as fast as the server produced them, for 60 seconds. Speed is seconds of video received divided by the seconds from the first segment to the last. Each server ran alone, with the other two paused, three runs per test.

Server Test Three runs Median speed Median time to first segment Median output bitrate
Plex a 2.65x, 2.70x, 2.79x 2.70x 1.43 s 4.62 Mbps
Plex b 8.17x, 8.35x, 8.45x 8.35x 0.84 s 2.35 Mbps
Plex c 9.15x, 9.32x, 9.36x 9.32x 0.83 s 3.14 Mbps
Jellyfin a 2.75x, 2.75x, 2.77x 2.75x 2.90 s 3.39 Mbps
Jellyfin b 9.46x, 9.51x, 9.68x 9.51x 0.96 s 1.81 Mbps
Jellyfin c 8.64x, 9.23x, 9.24x 9.23x 1.04 s 2.25 Mbps
Emby a 2.73x, 2.81x, 2.83x 2.81x 1.90 s 3.50 Mbps
Emby b 8.79x, 8.89x, 8.92x 8.89x 1.06 s 1.85 Mbps
Emby c 9.60x, 9.72x, 9.81x 9.72x 0.95 s 2.34 Mbps

Ratios to the fastest median on each test (a, b, c) and speed points (7 x their geometric mean):

  • Plex: 0.960, 0.878, 0.959; geometric mean 0.932; 6.52 points.
  • Jellyfin: 0.980, 1.000, 0.949; geometric mean 0.976; 6.83 points.
  • Emby: 1.000, 0.935, 1.000; geometric mean 0.978; 6.85 points.

All three encoded with x264 at its "veryfast" preset, with a quality setting (CRF) of 23, except Plex, which used 21 on test b and 22 on test c. Every output was H.264 at the requested size (1920x1080 on a, 1280x544 or 1276x544 on b, 1920x800 on c, keeping each film's shape), and each server's own log shows the English subtitle track drawn into the picture on test b. Output rate stayed steady through each 60-second run, with no pauses.

How close the call is

Jellyfin leads Emby by 1.25 points and Plex by 1.48 (8.40, 7.15 and 6.93 before rounding). Emby leads Plex by only 0.23.

  • Plex first: moving 9 of the 100 weight points from Cost (or from Setup and privacy) to Watching away from home puts Plex ahead of Jellyfin, 7.67 to 7.59 for the move from Cost. Eight points is not enough.

  • Plex second: moving 2 points from Setup and privacy to Watching away from home, or 3 from Cost, puts Plex ahead of Emby.

  • Emby above Jellyfin: it takes 18 points moved from Cost to Watching away from home for Emby to pass Jellyfin (6.79 to 6.78), and by then Plex is far ahead (8.42). No shift of this kind puts Emby first.

  • Leaving cost out: for a reader who will pay whatever it costs, dropping Cost and scaling the other weights up to 100 gives Plex 8.67, Jellyfin 7.87 and Emby 7.53.

  • Paying for a domain: Cost counts Jellyfin's and Emby's remote streaming as free, but the encrypted route that works on TV apps (a reverse proxy) needs a domain name, about $10 to $15 a year. Counting $60 over five years gives Jellyfin a Cost score of 8.0 and a total of 7.9, still first.

That is why the page names Jellyfin best overall but tells anyone who shares outside the home and will pay that Plex is the pick for them.

What changed after the rubric was written

  • The rubric's opening paragraph said Plex could not fetch metadata without signing in and that each server's setup had been run. Both were wrong (see the note under the rubric); we added a dated note and kept the original wording. Not a scoring change.
  • The library count in the rubric said 61 files (14 episodes); the manifest, fixed before any library was added, has 60 (13). Corrected on October 28 before any result; the correction is not a separate entry in our version history, which has the rubric (22:20 UTC) and the first full draft (23:20 UTC).
  • The right answer for the Battlestar Galactica (2003) episode first accepted only the 2004 series. A file labeled 2003 fits the 2003 miniseries just as well, so after the first results we accepted either. That changed Plex and Jellyfin from wrong to right on that file. With the original answer key the counts are Plex 59, Jellyfin 58 and Emby 56, which would score 9.8, 9.7 and 9.3 and change no position.
  • Plex's early count was 58, not 59. Besides the Battlestar Galactica key, our checking script kept only the first episode Plex listed for the two-episode Friends file, and in the early run that was episode 17, so the file counted as wrong. Plex lists the file under both episodes 16 and 17. We changed the script to look for episode 16 among all of a file's listings and to record every listing, then re-ran it for all three servers: Plex 60, Jellyfin 59, Emby 56, the same as before the change.
  • The rubric says throttling was turned off. Jellyfin's is off by default; Plex's and Emby's settings were left at their defaults, and every run produced video at a steady rate for the full 60 seconds, so no throttle engaged.

Other evidence, not scored

In the five minutes after a restart, with no one using it, each server looked up these names on the internet (recorded with tcpdump on the test machine, October 28, 2026). Plex, never signed in: servers.plex.tv, analytics.plex.tv, meta.plex.tv, metadata.provider.plex.tv, resources-cdn.plexapp.com and a plex.direct address. Jellyfin: repo.jellyfin.org and one of its download mirrors. Emby: www.mb3admin.com (Emby's back end), api.ipify.org (a public IP address lookup) and api.github.com. The connections are encrypted, so this shows which services each server contacts, not what it sends. The privacy point in Setup and privacy is scored from what each vendor says, as the rubric sets out.