Factorio 2.1 Experimental Server Guide: Opt-In, Mods & Rollback

Factorio 2.1 Experimental on Your Server — Opt-In, Mods, and Rollback

Last verified: September 20, 2026 against Factorio's own release API: 2.0.77 is stable for every build including headless, and 2.1.19 is the current experimental. Wube has described 2.1 as the final major update, but do not treat that as a stable-release date.

Factorio 2.1 is an opt-in experimental branch alongside stable 2.0.77. Wube has described 2.1 as the final major update, but the practical server decision today is simpler: keep a production world on stable until the group, save, and mod list have been tested against the current experimental build. This guide explains how to pin the branch, protect a save, and decide whether your multiplayer server should move.

What 2.1 is (and deliberately is not)

Wube has scoped 2.1 as quality of life, small features, polish, and modding improvements — explicitly ruling out new planets, enemies, research trees, and resource chains. The meaty announced area is space logistics (Friday Facts #441):

• Rocket silos will request exactly the number of items a platform actually needs for construction, ending the over-shipping dance.
• Platforms will prioritize foundation and cargo bays before other construction materials.
• Platforms can request items from all planets by default, with an optional filter to pin a request to one planet.

For multiplayer Space Age saves, that last one quietly removes an entire genre of "why is the platform starving" support questions. Community wishlists (visible planets from platforms, interplanetary circuit connections) are circulating, but treat anything not in an FFF as speculation — the full announced picture lives in our Factorio 2026 roadmap, updated as FFFs land.

How experimental works for a server

• It is opt-in on both ends. Experimental builds publish alongside stable: a Steam beta branch for clients, and experimental-tagged headless packages from factorio.com for servers. Nothing moves until you move it.
• Version match is absolute. Clients and server must run the same version. Moving a community server to 2.1 experimental means asking every player to flip their branch — which is why most multiplayer servers should not be first-week adopters unless the group explicitly wants to be.
• Experimental iterates fast. During an experimental cycle, point releases can land multiple times a week. Headless servers either chase them (auto-update + announced restarts) or pin one build and update deliberately. Pin; chasing is for test servers.

What the 2.1.19 headless package actually contains

Announcements are one thing; here is what is inside the experimental headless build itself, read out of the package rather than from a roadmap post. This is 2.1.19, internal build 87445, linux64, headless, with a map output version of 2.1.19-0 against stable 2.0.77's 2.0.77-0. That pair of strings is the no-downgrade rule made concrete, and it is the thing to compare when you are unsure whether two installs will read the same save.

The mod list grows from five to six

Stable 2.0.77 ships five bundled mods: core, base, elevated-rails, quality and space-age. 2.1.19 ships six. Recycling moved out of the quality mod into a new bundled mod called recycler, and the dependency lines record the move: on 2.1.19 quality declares base >= 2.1.0, recycler >= 2.1.0, where on 2.0.77 it declared only base >= 2.0.0.

Three consequences for a server:

• A hand-written mod-list.json will be missing an entry. If you maintain that file yourself rather than letting the game write it, recycler is the line you will not have.
• Any mod that touched recycling prototypes is touching a different mod now. This is a concrete reason to put recycling-adjacent mods at the top of the audit list below rather than treating the whole list as equally risky.
• The bundled mods gained a new marker. On 2.1.19 the four expansion mods, elevated-rails, quality, recycler and space-age, all declare expansion_required, a field none of them carries on 2.0.77.

The base game is nearly untouched, and that is the point

Comparing the two builds with only base enabled, 2.1.19 resolves 217 recipes, 196 technologies, one planet and 245 items, against 2.0.77's 217 recipes, 196 technologies, one planet and 244 items. Identical recipe and technology counts, and a single new item. Wube said 2.1 rules out new research trees and resource chains, and the shipped data agrees with them. If your server is vanilla, the content risk in moving is genuinely small; the risk is in the mods and in the one-way save.

Four new launch arguments

The option parser prints its own registry, so the argument list is comparable between builds directly. 2.1.19's registry names four arguments 2.0.77's does not:

--exchange-string is the useful one for anybody starting a server. It takes the map exchange string you copy out of the map generator screen and applies it directly at world creation, which removes the whole step of translating an exchange string into a hand-written map-gen-settings.json. Note the ordering the parser states: the exchange string is loaded first, then any other settings arguments are processed on top, so you can use one as a base and override a single value.

The network side is unchanged. The Server argument group is the same thirteen arguments in both builds, the same 53 console commands, the same 19 /config settings and the same 28 documented server-settings.json keys. Nothing you currently pass to a 2.0.77 server needs rewriting for 2.1.19. See port configuration for that list and server settings for the file.

The no-downgrade rule

The single most expensive mistake of every experimental cycle: saves do not load backwards. A map touched by 2.1 will not open under 2.0. The binary downgrades fine; your two-hundred-hour map does not. So:

• Copy the save file itself before switching — a real file copy you control, not only a panel snapshot you have never restored.
• Keep the 2.0 backup until the group commits. If 2.1 experimental goes badly (a mod you depend on lags, a bug bites your base), the rollback is: restore binary, restore save, lose the sessions played on 2.1. Decide in advance whether that trade is acceptable, and tell your players.

Mods: audit before you flip

2.1 concentrates changes in space-platform logic and the modding API — exactly where mods break. The pattern from every previous cycle: major mods update within days, mid-size mods within weeks, abandoned mods never. Practical sequence: export your mod list, check each mod's portal page for 2.1 releases after experimental lands (the portal serves per-version builds — your server on 2.1 pulls 2.1 releases, which will be young), and classify each as critical / nice-to-have / droppable. A server whose permissions or protection mod is "droppable" has a different definition of droppable after the first griefing. Our mod installation guide covers the per-version mechanics.

2.1 experimental is where the ARM64 Linux build lives, headless included

Friday Facts #446, 18 September 2026, shipped something a host cares about more than the space logistics work: "Standalone Factorio for ARM64 Linux is available on our website, including the headless version." On Steam it is not on stable. Wube's own instruction is that "Factorio for ARM64 Linux is available on the experimental branch (version 2.1), so please opt into experimental if you are playing Factorio on the Steam Frame or other ARM64 Linux devices". So an ARM box, a Raspberry Pi or an Asahi Linux MacBook can run the server, and on Steam that means living on the experimental branch with everything below about mods and rollback still applying.

On whether ARM and x86 players can share a world, the only evidence Wube has published is their own testing: "we never experienced an ARM64 vs. x86 related desync" across the May 2026 LAN party, credited to the earlier Nintendo Switch and Apple Silicon ports. That is a report of an absence, not a compatibility guarantee, and the post makes no statement at all about save files moving between architectures. Our own nodes are AMD x86, so nothing here changes a server you rent from us.

The decision matrix

• Fresh group starting now: choose stable 2.0.77 for a predictable production world, or choose 2.1 experimental only if everyone understands the update cadence and mod risk. Dial in map-exchange settings with the Factorio map generator before world creation.
• Active long-running save, vanilla: wait for two or three experimental point releases, then move with a verified backup. The space-logistics improvements are worth it; the first-week bugs are not.
• Active save, heavily modded: wait for stable, or until every critical mod publishes and survives testing on the 2.1 build you intend to run.
• Archival megabase: no rush, ever. 2.1 is the final destination; it will still be there when every mod has caught up. The update mechanics are in how to update your server.

Want the version flip to be a dropdown instead of a package swap? Host a Factorio server with Supercraft — pin 2.0, test 2.1 experimental, and move when your group is ready.

Looking for managed Factorio server hosting? Supercraft runs Factorio dedicated servers with daily backups, instant setup, and 4 region options.