BlueClient Download

Benchmarks

Every speed number BlueClient gives, measured on one PC: five runs of each client, taken in turns, the middle one shown with the slowest and fastest beside it. The slow results are here too. Measured on 7 and 8 October 2026.

A draw with plain Sodium in an open field
Flying over empty land, the clients built on Sodium finish close together: BlueClient 824, Fabric with Sodium alone 791, and the same 1% low. There, Sodium does most of the work.
Ahead in a crowd
A plaza with 160 mobs and armour stands: 444 frames a second against 353 for Sodium alone. With 1,000 armour stands: 71 against 54.
Vulkan on 26.2
On 26.2 and newer BlueClient draws with Vulkan when shaders are off: 963 frames a second against 806 with OpenGL, and the 1% low from 296 to 380.
First to the title screen, since 1.14
10.9 seconds from Play to the title screen on 26.2, against 16.9 for Sodium alone and 17.3 for Fabric alone. Until 1.13.7 BlueClient was the slowest of the three, and those rows stay on this page.
Lunar and Feather
Not measured here. Both start the game from their own launcher with their own sign-in, which this test cannot do. The world and the steps under Check it yourself work in any client.

Play to title screen

Seconds from the game starting to the title screen, loading screen gone. Lower is better. Measured on 8 October with the bench's limit for other programs at 30 % instead of 15 (a busy evening on the PC), the same for every client.

Play to title screen, Minecraft 26.2
Minecraft 26.2, BlueClient 1.14seconds to the title screenruns kept
BlueClient10.89 10.77–10.985/5
Fabric + Sodium16.93 16.68–18.185/5
Fabric alone17.28 17.00–17.545/5
Play to title screen, Minecraft 1.21.11
Minecraft 1.21.11, BlueClient 1.14seconds to the title screenruns kept
BlueClient14.24 13.79–20.055/5
Fabric + Sodium22.55 22.00–28.545/5
Fabric alone22.09 21.86–27.275/5

Most of this lead is one request. A game that is still loading has no window yet, so Windows treats it as background work, and on this laptop's processor that means the slower cores at a low speed. BlueClient's in-game mod asks Windows not to do that, and the loading that follows takes about a second where it took seven. Fabric alone, given the same request from outside, reached the title in 11.0 seconds instead of about 17: the request is the difference, and any client could make it. The rest is the Mojang screen's goodbye, 0.8 seconds where the game takes 2. The other clients here were started the same way as BlueClient, not from their own launchers, which may or may not make the same request; that is not measured here.

Before 1.14: the slowest of the three

Measured on 7 October with BlueClient 1.13.7. These rows stay here.

Play to title screen with BlueClient 1.13.7, Minecraft 26.2
Minecraft 26.2, BlueClient 1.13.7seconds to the title screenruns kept
BlueClient18.21 17.63–20.716
Fabric + Sodium16.56 16.32–17.597
Fabric alone16.77 16.45–17.2111
Play to title screen with BlueClient 1.13.7, Minecraft 1.21.11
Minecraft 1.21.11, BlueClient 1.13.7seconds to the title screenruns kept
BlueClient24.47 19.53–25.115/5
Fabric + Sodium22.18 21.74–27.245/5
Fabric alone21.58 20.82–27.675/5

The fixed lap

A flight over Minecraft's own terrain: spectator mode, a square 160 blocks a side at height 130, one lap every 64 seconds, new chunks coming into view the whole time. The world flies it by itself, the same in every client.

Frames per second on the fixed lap, Minecraft 26.2
Minecraft 26.2average fps1% lowslowest 1% under (ms)longest frame (ms)garbage collection (ms a second)memory in use (MB)
BlueClient824 803–830300 266–3332.29 2.18–2.5722.8 21.1–33.32.0 1.9–2.81265 1232–1368
Fabric + Sodium791 772–802297 293–3252.28 2.19–2.3525.2 22.6–30.52.6 2.4–3.41276 1269–1352
Fabric alone464 452–467117 115–1244.89 4.73–5.0951.1 28.0–90.65.0 4.9–5.01359 1344–1429

This is a draw between the clients built on Sodium: BlueClient is 4 % ahead of Fabric with Sodium alone, with the same 1% low.

Frames per second on the fixed lap, Minecraft 1.21.11
Minecraft 1.21.11average fps1% lowslowest 1% under (ms)longest frame (ms)garbage collection (ms a second)memory in use (MB)
BlueClient814 798–870293 265–3262.39 2.20–2.6921.2 20.4–33.22.2 2.0–4.31280 1219–1353
Fabric + Sodium805 787–850305 295–3162.25 2.15–2.3821.3 20.2–29.12.3 2.3–4.31233 1070–1312
Fabric aloneNot measured. On all five runs the game's log did not name the graphics card it drew on, and the bench refuses a run it cannot place on the right card.

On 1.21.11 it is a draw again: 814 against 805, and Sodium alone has the better 1% low (305 against 293).

Crowds

A lit plaza with 100 armour stands in armour and 60 zombies, skeletons and villagers, the camera circling them. Then the same plaza with 1,000 armour stands.

The hub: 100 armour stands and 60 mobs, Minecraft 26.2
Hub, 160 entitiesaverage fps1% lowslowest 1% under (ms)garbage collection (ms a second)
BlueClient444 431–453195 184–2033.93 3.78–4.092.7 2.7–2.8
BlueClient without its in-game mod433 425–441195 178–2054.06 3.80–4.352.7 2.5–2.9
Fabric + Sodium353 349–357164 149–1664.67 4.59–4.782.9 2.7–3.3
The crowded hub: 1,000 armour stands, Minecraft 26.2
Hub, 1,000 armour standsaverage fps1% lowslowest 1% under (ms)garbage collection (ms a second)
BlueClient71 68–7345 41–4618.60 17.18–21.323.4 3.2–3.5
BlueClient without its in-game mod63 62–6838 36–4321.58 19.33–23.993.4 3.3–3.6
Fabric + Sodium54 52–5534 33–3625.51 24.85–25.663.8 3.3–3.9

Most of the lead over Sodium comes from the mods BlueClient adds around it. BlueClient's own in-game mod makes little difference with 160 entities (444 with it, 433 without) and more with 1,000 (71 against 63).

Vulkan against OpenGL

The fixed lap on 26.2, each client drawing with OpenGL and then with Vulkan, the newer way Minecraft can talk to the graphics card. BlueClient picks Vulkan by itself on 26.2 and newer when shaders are off; shaders still need OpenGL.

Vulkan against OpenGL on the fixed lap, Minecraft 26.2
Lap, 26.2average fps1% lowslowest 1% under (ms)garbage collection (ms a second)
Fabric alone, OpenGL456 432–457119 113–1204.93 4.91–5.284.8 4.3–5.0
Fabric alone, Vulkan597 547–608167 151–1733.60 3.44–4.408.6 7.5–9.0
Fabric + Sodium, OpenGL798 782–801300 289–3142.34 2.24–2.372.5 2.4–3.3
Fabric + Sodium, Vulkan975 957–985389 359–4151.76 1.73–1.972.5 2.4–3.2
BlueClient, OpenGL806 787–841296 262–3112.21 2.20–2.652.4 2.0–2.9
BlueClient, Vulkan963 937–979380 336–4311.79 1.71–2.102.2 1.7–3.2

What Iris costs with shaders off

Iris is what runs shader packs. BlueClient keeps it installed so shaders are one switch away, and it is not free even when they are off.

What Iris costs with shaders off, the hub, Minecraft 26.2
Hub, shaders offaverage fps1% lowslowest 1% under (ms)
BlueClient (Iris installed, shaders off)447 437–452185 178–2003.93 3.80–4.02
BlueClient without Iris465 450–481201 184–2053.73 3.65–3.88

About 4 % of the average and 8 % of the 1% low in the plaza.

Faster ticks for mobs and particles

Twenty times a second the game moves every mob and particle, and the frames that carry that work are the slow ones. BlueBoost answers two of the game's collision questions from memory when nothing around them has changed — the same answers, checked bit for bit. The scene is a working farm: ten cows dying and ten more appearing every two seconds, 64 campfires smoking.

Faster mob and particle ticks, on and off, the working farm, Minecraft 26.2
Working farmaverage fps1% lowslowest 1% under (ms)longest frame (ms)frame with a tick (ms)frame without (ms)
Off, run 13171107.4849.56.652.92
Off, run 1 again308958.3442.96.942.99
On, run 12951296.3416.55.273.25
Off, run 23031167.1026.36.293.09
On, run 22311077.8119.66.804.09
Off, run 3217859.4039.08.094.26
On, run 33451555.2816.44.752.79
Off, median3061027.9141.06.803.04
On, median2951296.3416.55.273.25

With it the 1% low went from 102 to 129 and the longest frame from 41 to 17 ms; the average did not move beyond the noise. This one was measured while another copy of Minecraft was open on the PC, so it is noisier than the other tables: run by run the numbers wander, and the medians are the result.

Frame generation on a slow graphics chip

When the graphics chip is the limit and a lower resolution is not enough, BlueBoost can show one extra picture between every two the game draws, made from the two around it. The hotbar and text are always the game's own. It never turns on with a fast card.

Frame generation on a slow graphics chip (simulated), Minecraft 26.2
Farm, 1080p, turningframes the game draws a secondframes shown a second
FPS Boost off21.021.0
FPS Boost Auto, lower resolution36.736.7
FPS Boost Auto, lower resolution + frame generation35.070.1

Measured on 26.2's farm at 1080p with the camera turning, the RTX held back by a test load to the speed of a weak laptop chip — not on a real built-in chip, which we have not been able to test yet. The extra pictures add about 14 ms of delay (half of one real frame), and the in-game FPS counter keeps showing the frames the game draws.

How we measure

The PC

A laptop: Intel Core i5-14500HX, NVIDIA GeForce RTX 5050 Laptop GPU (driver 32.0.16.1742), 24 GB of memory, Windows 11, plugged in. Its built-in Intel graphics is not used, and a run that ends up on it is thrown away.

The game

Minecraft 26.2 and 1.21.11 on Fabric Loader 0.19.5, in a 1920 × 1080 window: render distance 16, simulation distance 12, Fancy, V-Sync off, no frame limit. Java is the one Mojang ships for that Minecraft (25 for 26.2, 21 for 1.21.11) with 8 GB of memory at most and 2 to start.

The same start for everyone

Every client is the same game started with the same command line, the one BlueClient's launcher builds. Only the mods folder changes.

Five runs, taken in turns

Every client runs once, then every client again, five times over, so a laptop warming up through the hour slows them all alike. Each table shows the middle run of five; the small numbers beside it are the slowest and the fastest. When those ranges overlap, the difference is not a result.

Every frame timed

A tiny test mod notes the moment every frame starts. The first 15 seconds in the world are loading and left out; the next 60 are the result. Average fps is frames divided by seconds. 1% low takes the slowest 1 % of frames, averages their length and turns it back into frames a second: what a stutter feels like. Slowest 1 % under: 99 frames in 100 were quicker than this. Garbage collection is how long Java paused the game each second to free memory.

A quiet PC, or no result

Before each run the test waits until no other Minecraft or build is running and the processor is mostly idle. During the run it watches every other program: if they used more than 15 % of the processor between them, or another Java program used 2 %, the run is thrown away and run again, twice at most. The tables say how many runs were kept.

Play to title

Seconds from Java starting to the title screen with the loading screen gone. The launcher's own work before that is not in it (BlueClient's is about 0.2 seconds). Never the first start after the PC is switched on: that one takes about twice as long for every client, because Windows has nothing in memory yet.

The clients

BlueClient: what its launcher installs — its in-game mod and BlueBoost, with Sodium, Lithium, FerriteCore, Krypton, Iris (shaders off), Entity Culling, LambDynamicLights, ImmediatelyFast, More Culling, BadOptimizations, No Chat Reports, Cloth Config and Fabric API. Fabric + Sodium: Fabric API and Sodium only, the base most performance clients are built on. Fabric alone: Fabric API only, the game's own renderer.

Dynamic FPS is left out of every test, because it only acts when the game is in the background, and so is McPvP.com Tier Tagger, which a new BlueClient profile also gets and which only shows PvP tiers beside names.

The lap, crowd, Vulkan and Iris tables were measured with BlueClient's mods as installed on 6 October 2026 (in-game mod 1.57.2, BlueBoost 0.1.0); the tick, frame generation and Play to title tables with BlueBoost 0.2.0 and the in-game mod 1.57.3.

Check it yourself

The lap is a normal Minecraft world. It works in any client, ours or anyone else's.

With any client, by hand

  1. Download the lap world for Minecraft 26.2 or 1.21.11 (33 MB each).
  2. Unzip it into that client's saves folder for the same Minecraft.
  3. Set the window to 1920 × 1080, shaders off, render distance 16, simulation distance 12, V-Sync off, no frame limit.
  4. Open the world. It puts you in spectator mode and flies the lap by itself.
  5. From 15 seconds in, record 60 seconds with CapFrameX or PresentMon and note the average and the 1% low.
  6. Do it five times, taking turns with the client you compare it to, and take the middle result.

The zip also holds the plaza, the crowded plaza and the farm: its read-me says how to switch the world to each.

With our test, exactly

The test is a few scripts in BlueClient's source, in tools/bench. These are the commands behind the tables on this page:

node tools/bench/build-probe.js --mc=26.2
electron tools/bench/install.js
node tools/bench/make-worlds.js
node tools/bench-fps.js --runs=5
node tools/bench-fps.js --runs=5 --mc=1.21.11
node tools/bench/bench.js blueclient no-companion sodium-only --world=hub --runs=5
node tools/bench/bench.js blueclient no-companion sodium-only --world=hub1000 --runs=5
node tools/bench/bench.js vanilla vanilla+vk sodium-only sodium-only+vk blueclient blueclient+vk --world=route --runs=5
node tools/bench/bench.js blueclient no-iris --world=hub --runs=5
node tools/bench/bench.js blueclient sodium-only vanilla --title --runs=5
node tools/bench/page-tables.js

The last one writes the tables on this page from the results.