Sodium, Iris and Fabric on Minecraft 26.3: what works, what breaks

Every Minecraft update produces the same wave of searches: does my shader pack still work, is Sodium updated, did they break my farm. Minecraft 26.3, "Wilderness Bound," released 15 September 2026, and this page exists to answer those questions honestly, which means separating three very different kinds of claim.

Short answer: yes, the core performance and shader mods work on 26.3. Sodium, Iris, Lithium and Fabric API all published stable 26.3 builds within a day of release. The table below has the exact versions, checked against Modrinth on 29 September 2026.

Mod compatibility, checked against Modrinth directly

Rather than guess, we queried the Modrinth version API for each project and read the actual game_versions and release channel of every 26.3 build. That is what the mod authors have published, not a prediction.

What that means in practice:

• Fact: Sodium and Iris shipped stable 26.3 builds on release day, 15 September. Lithium followed on 16 September and got another update on 28 September. Fabric API tracked the snapshots throughout and has stable 26.3 releases.
• Fact: Sodium also has a newer 0.9.3 alpha for 26.3 (20 September). Unless you want to test it, stay on the stable 0.9.2.
• Fact: Iris for 26.3 was Fabric only at first; since 1.11.7 (30 September) it has a NeoForge build as well.
• What to check yourself: everything else in your mod list. The mods above are the common performance core; your specific content mods, minimaps and utilities each need their own 26.3 build. Check each mod's Modrinth or CurseForge page for a 26.3 tag before updating.

If you're not sure how your current mod stack fits together, the complete Fabric performance mod stack and Sodium vs Embeddium cover the current, working baseline you'd be updating from.

The rendering change: OIT replaces Improved Transparency

This is the technical change most likely to have downstream effects, and it's worth being precise about what it actually is.

Snapshot 2 of 26.3 replaced the old Improved Transparency video setting with Order-Independent Transparency (OIT), a different algorithm for compositing translucent geometry: stained glass, water, ice, and leaves layered on top of each other. The old system's flickering and incorrect sort order on translucent surfaces is exactly what OIT targets.

• Fact: this is a change to how the vanilla renderer composites transparent pixels, confirmed in the Snapshot 2 notes and still present as of Pre-Release 1. Pre-Release 1 also added core/blit_depth.fsh, an alternative depth-copy path for the OIT renderer, for devices where the existing backend approach for copying depth does not work. That is a compatibility fix for OIT itself, not a sign the feature is still in flux.
• Reasoning we're labeling as such: any mod or shader pack that hooks into the transparency pass, reorders render layers, or assumes the old Improved Transparency pipeline is doing the compositing has a plausible reason to need updating. Iris in particular sits directly in this part of the render pipeline, which is one reason shader compatibility is worth watching closely here, not just assuming will carry over.
• What we don't know: whether OIT breaks any specific shader pack. Iris now has a stable 26.3 build (1.11.7), so packs can be run against the new renderer, but we have not tested individual packs on it. If a pack looks wrong on 26.3, especially glass, water or particles, update the pack itself first; the transparency pass is the part most likely to need a shader-side fix.

This sits in the same direction as 26.2's experimental Vulkan renderer: Mojang is modernizing the rendering pipeline in stages rather than one rewrite. If you track how these rendering changes interact release over release, that guide is the place to watch.

Datapacks and map-making: format bumps are the real breakage here

Unlike client mods, this is territory where the snapshots and pre-release notes themselves hand you a concrete, verifiable answer: data pack format has moved from 108.0 (Snapshot 1) to 121.0 (Pre-Release 3, and unchanged through both release candidates), and resource pack format from 89.0 to 97.1, over the course of the cycle. The data pack number moved four times late in the cycle: 118.0 at Snapshot 10, 119.0 at Pre-Release 1, 120.0 at Pre-Release 2, and 121.0 at Pre-Release 3. It then held steady at 121.0 through Release Candidate 1 and Release Candidate 2, which is a genuine signal the number is now settled ahead of release. If you are tracking any number older than 121.0, it is already stale. A datapack or resource pack built for 1.21.4 or 26.2 will not load cleanly on 26.3 without its pack.mcmeta format number updated, and depending on what it uses, more than the number.

Concrete, verified changes that hit datapacks directly:

• New block predicate minecraft:height_range, usable in loot tables and other predicate contexts.
• New feature types (end_podium, overlay, projected_random_patchy_square), alongside renamed existing ones (glowstone_blob became random_neighbor_spread, basalt_pillar became single_block_pillar in Snapshot 2). A datapack referencing the old feature type names by their pre-26.3 identifiers will not resolve.
• New data components, including minecraft:block_transformer, minecraft:brewing_fuel, minecraft:cooking_fuel, minecraft:compostable, minecraft:villager_food, and minecraft:mob_visibility. None of these retroactively break an existing item definition, but they are new territory for anyone building custom items or loot.
• Material Rule handling moved out of noise settings into its own data-driven registry, and carver configuration was inlined rather than nested under a config key. Both are structural changes to world-generation datapacks specifically, not to loot tables or recipes.
• Trim material format overhaul: asset_name was replaced with palette_id. Any custom armor trim datapack using the old field name needs updating.
• The two new commands, /compute and /posteffect, expand what a datapack can do (/compute evaluates a number provider for scoreboard-style math; /posteffect manages post-processing shader effects on a player) but don't break anything by existing. For the day-to-day mechanics of installing and troubleshooting datapacks generally, see how to install datapacks.

Pre-Release 1 (1 September 2026) added three more changes datapack authors need to know about:

• The number provider system was split into separate integer and float registries, with new maths operations available in each. /compute and /data compute were both adjusted to match the split, so a datapack calling either command with a number provider written against the pre-split system should be re-checked.
• cookingtime is now a mandatory field in cooking recipes (furnace, smoker, blast furnace, campfire) rather than optional. A recipe JSON that omitted it and relied on a default will no longer load.
• Block state providers can now be standalone registry types, referenced by string rather than always inlined, which gives world-generation datapacks a shorter way to reuse the same provider in multiple places.

Pre-Release 2 (4 September 2026) then moved the data pack format again, to 120.0, and renamed a structure tag:

• on_abandoned_camp_windswept became on_abandoned_camp_windswept_forest. A datapack referencing the old tag name will not resolve it. This is a small rename with an outsized consequence, because a tag that fails to resolve is a silent failure rather than a loud one.
• Sculk sensors now detect ice and snow melting. Not a format change, but worth knowing if you build sculk-based contraptions or maps: a sensor near melting ice now has a vibration source it did not have before.
• Resource pack format stayed at 97.1.

Pre-Release 3 (8 September 2026) moved the data pack format again, to 121.0 (resource pack format held at 97.1), and changed two float number providers for consistency with their int equivalents:

• minecraft:mod now uses standard modulus instead of floor modulus. A datapack relying on floor-modulus behavior from this float provider (for example, expecting a negative result to come out positive) will get different output.
• minecraft:pow now errors and halts computation when raising 0 to the power of 0, instead of silently continuing. A datapack that can hit this case needs to guard against it or expect the function to stop.
• 19 bugs were fixed. This was the last pre-release before 26.3 moved into the release candidate phase.

Release Candidate 1 (10 September 2026) kept data pack format 121.0 and resource pack format 97.1, and fixed four things: weapon components no longer disable blocking items, axes no longer disable shields, the match_block loot predicate no longer always requires a block entity, and an ender pearl block-transparency visual glitch. The match_block fix is the one datapack authors should care about: a loot table using that predicate against a blockless target should be re-tested, since the old behavior may have been masking a false negative.

Release Candidate 2 (11 September 2026) also kept both format numbers unchanged, with a single fix unrelated to datapacks: clicking a button that opens a folder no longer freezes the game.

If you maintain a datapack, the format-number bump alone is reason enough to test against 26.3 rather than assuming your pack loads, since Mojang has already told you, in the pack format numbers, that something changed. The number held at 121.0 across Pre-Release 3, both release candidates and the final release, so 121.0 is what you target now.

Farms: what we can say, and the Drowned lesson

Farm guides are exactly where "wait for release" earns its caution, and 26.3 already produced the textbook example of why.

Fact: Snapshot 2 changed Drowned so that one holding a trident within 3 blocks of a target would switch to melee instead of throwing. Fact: Snapshot 10 (25 August 2026) reverted that change, and it held through Release Candidate 2 and into the 15 September 2026 release. Drowned are back to their original trident-throwing behavior, confirmed at release. No reason was given for either the change or the revert.

That reversal, mid-cycle, is the whole argument for not rewriting a Drowned farm's mechanics based on a snapshot. A mechanic can ship, hold for eight snapshots, and still get pulled before release, which is exactly what happened here. Nothing since Pre-Release 1 carried a further mob-behavior change, for Drowned or anything else (iron golems, guardians, endermen, piglins, and the rest remain unchanged), and that held all the way to release.

The wider farm risk in 26.3 is not mob AI, it's structural: none of the 26.3 content (dappled forest, abandoned camps, poplar wood, straw beds, cushions) touches redstone mechanics, hopper behavior, or existing farm designs. Existing iron, gold, guardian, enderman, and mob-grinder designs from the current farm library carry over mechanically, confirmed against the shipped release. See the farm guide collection for the current designs this applies to.

Servers: what to actually do before updating

This is process advice, not a compatibility claim, and it doesn't change release to release.

1. Never update a live world's server jar on release day. Copy the world folder to a test environment first. World format changes are one-directional: Minecraft upgrades a world's data on load and does not reliably downgrade it.
2. Check plugin and mod compatibility before the players do. If you run Paper or a fork, wait for your plugins to publish 26.3-compatible builds the same way you'd wait for Sodium above; a plugin built against a prior API can throw errors or silently misbehave on a version it wasn't tested against. If you run Fabric, the same Modrinth-check method used above for Sodium and Lithium works for every server-side mod you depend on.
3. Back up before you touch anything, snapshot testing included. If you're unclear on your current backup routine, the server backup guide covers what to automate.
4. Budget for a compatibility gap. Based on past cycles, expect days to a couple of weeks between release and your specific plugin or mod stack catching up, not zero. Plan the update window around that, not around release day itself.
5. Re-test performance after updating, not just functionality. A new renderer or new block types can shift what your server profiles as slow; the complete performance guide and Aikar JVM flags are your baseline for re-tuning if TPS or memory behavior looks different after the jump.

For the update procedure itself, step by step for both a singleplayer world and a server, see how to update to Minecraft 26.3 safely.

FAQ

Does Sodium work on Minecraft 26.3?

Yes. Sodium 0.9.2 for 26.3 was released on 15 September 2026, the same day as Minecraft 26.3, for both Fabric and NeoForge. A 0.9.3 alpha also exists, but the stable 0.9.2 is the one to use.

Is Iris updated for 26.3?

Yes. Iris had a stable 26.3 build on release day (15 September 2026), and the current version is 1.11.7, for Fabric and NeoForge. Because 26.3 changed the transparency pipeline (OIT), update your shader pack too if glass or water looks wrong.

Do Sodium and Iris work together on 26.3?

Yes: install Sodium 0.9.2 and Iris 1.11.7 for 26.3, plus Fabric API on Fabric. Since 1.11.7, Iris also has a NeoForge build for 26.3.

What is OIT in Minecraft 26.3 and does it break shaders?

OIT (Order-Independent Transparency) is a new rendering method that replaced the old Improved Transparency setting, aimed at fixing flicker and sort-order bugs on translucent surfaces like glass and water. It's a real change to the render pipeline that mods and shaders touching transparency have a plausible reason to need updating for. Iris has a stable 26.3 build, so packs can run on it, but results depend on the pack: if transparency looks wrong, update the shader pack first.

Will my datapack work on 26.3 without changes?

Only if it doesn't use the renamed feature types, the old trim material field, the now-relocated Material Rule configuration, the number provider and cookingtime changes from Pre-Release 1, the on_abandoned_camp_windswept tag renamed in Pre-Release 2, or the minecraft:mod and minecraft:pow float provider changes from Pre-Release 3, and even then you need to bump the pack format number (121.0 for the data pack format as of Release Candidate 2, up from 108.0 at Snapshot 1). Test it against the latest release candidate or the release build before relying on it.

Do my farms still work in 26.3?

Yes. Nothing across the ten 26.3 snapshots, the pre-releases, or the release candidates changed redstone, hoppers, or the mob AI behind common farm designs, except one Drowned trident change that was itself reverted before release. Existing iron, gold, guardian, and mob-grinder farms keep working in the shipped 15 September 2026 release.

When should I update my server to 26.3?

Now that 26.3 "Wilderness Bound" has released, update once your specific plugins or mods confirm support, unless you're running vanilla with no plugins or mods. Copy your world to a test server first, and keep a backup regardless.

---

Sources & further reading:

• Minecraft Wiki - Java Edition 26.3
• Minecraft Wiki - Java Edition 26.3 Pre-Release 1
• Minecraft Wiki - Java Edition 26.3 Pre-Release 2
• Minecraft Wiki - Java Edition 26.3 Pre-Release 3
• Minecraft Wiki - Java Edition 26.3 Release Candidate 1
• Minecraft Wiki - Java Edition 26.3 Release Candidate 2
• Minecraft Wiki - Java Edition 26.3 Development versions
• Modrinth version API, checked 29 September 2026: Sodium, Lithium, Iris, Fabric API
• Minecraft 26.3: everything we know so far - SyntaxMine
• The complete Minecraft performance guide - SyntaxMine
• Minecraft farm guides - SyntaxMine