Baldur's Gate 3 did not run on Unreal Engine 5, Unity, or any licensed middleware. It runs on Divinity Engine 4.0, Larian's in-house RPG engine, iterated since 2011 specifically to make multiplayer, systemic CRPGs with full cinematics.
That choice looked unfashionable in 2023. Every sensible producer would have picked Unreal for hiring, outsourcing, and next-gen lighting. Larian refused, and then became the first game to win Game of the Year at all five major ceremonies — Golden Joysticks, Game Developers Choice, D.I.C.E., BAFTA, and The Game Awards.
The short answer: Divinity 4.0 is not a graphics engine that happens to do RPGs. It is a dungeon-master engine that happens to do graphics. Larian kept it because commercial roadmaps would never prioritize barrelmancy, shoving goblins off cliffs, talking to every squirrel, and rolling a d20 in co-op while a cinematic plays.
What Divinity Engine 4.0 actually is
Divinity Engine was developed by Belgian independent Larian Studios alongside its Divinity series, which Larian began after its founding in 1996. The version history is simple:
- 1.0 for Divinity: Original Sin (2014): built on a few million dollars, shipped with a Kickstarter backer editor and even a joke Cow Simulator to prove moddability.
-
2.0 / 3.0 for Divinity: Original Sin II (2017): focused on better tools, documentation, and a shipped
DivinityEngine2.exeeditor. - 4.0 for Baldur's Gate 3: rebuilt around one hard problem — full performance-capture cinematics in multiplayer with a D&D 5e ruleset.
CEO Swen Vincke put it bluntly in 2020: “We've been working on the Divinity Engine since 2011. It's ever evolving and it's dedicated to making RPGs and multiplayer. The big thing for the [latest version] is the ability to do work cinematically.” That cinematic load was an estimated 1.5 million words of performance capture, with two motion-capture studios running continuously.
Technically, 4.0 added a Vulkan backend offered alongside DirectX 11 in Baldur's Gate 3, a pragmatic choice for a CPU-heavy systemic game rather than a marketing showcase for ray tracing. The game itself adapts D&D 5th Edition, uses turn-based combat with d20 rolls, advantage/disadvantage, and free real-time exploration outside combat. It launched in Early Access with Act 1 in August 2020 and hit 1.0 on Windows on 3 August 2023, followed by PlayStation 5, macOS and Xbox Series X/S later that year.
Why not just license Unreal? Larian developers later said the quiet part out loud: with a commercial engine, “the engine's roadmap is not necessarily your roadmap”. That line, expanded in Vincke's 2024 GDC-season interview about staying independent, is the whole thesis. No external product manager would greenlight a year of work so you can pick up explosive barrels in co-op without desync.
What the boring engine gained
1. Permission to say yes to chaos.
Original Sin II already let players combine fire, oil, water, electricity, and high ground. Baldur's Gate 3 ported that philosophy to D&D: shove, throw, dip, hide, jump as bonus action, speak with animals, speak with dead. Because Larian owned physics, save logic, and scripting (its Osiris-style rule system from Original Sin), a designer could add a stupid interaction without filing a feature request to Epic.
That is why the game feels like a real dungeon master. Most engines optimize for predictable content pipelines. Divinity 4.0 optimizes for improvisation.
2. Cinematics without giving up simulation.
The hardest technical bet was not lighting. It was building a powerful engine capable of full cinematics in multiplayer with complex rulesets. Single-player cinematics are easy. Four-player co-op where anyone can start dialogue, leave, attack the speaker, or polymorph mid-sentence is not.
Larian solved it by keeping simulation and presentation in the same house. The camera can swoop low because the engine knows who rolled what. Failure still gets a close-up. That matters more than it sounds — see my take below.
3. A Vulkan escape hatch for a CPU-bound game.
Baldur's Gate 3 is not a twitch shooter. Its bottlenecks are AI, pathing, visibility, and hundreds of interactable items per map. Offering Vulkan next to DX11 gave low-overhead options on varied PCs without rewriting the renderer for DX12. It was unglamorous, stable, and exactly what a 100-hour systemic RPG needs.
What it cost
Owning an engine is a tax most indies underestimate, and Larian paid it publicly.
Tools were for internal use. Vincke told PC Gamer in March 2024: “We are a game development company, we're not a tools company. So the tools are really for internal use.” The Original Sin II editor was powerful but, in his words, “a bit shaky” and crash-prone. Few full campaigns emerged despite Bethesda-Creation-Kit ambitions.
Baldur's Gate 3 repeated the pattern. Official mod support in Patch 7 (September 2024) deliberately limited the toolkit to classes, items, balance, NPC tweaks and cosmetic mods, without official level-editor support, partly because console mod browsing with Sony and Microsoft approval made full tools “complicated.”
The community immediately routed around it. Two days after Patch 7, modder Siegfre released BG3 Toolkit Unlocked, re-enabling disabled features including a full level editor. The shipped toolkit, in other words, was essentially a locked Divinity 4.0 editor.
Other costs were less visible: performance-capture localization that Vincke admitted could add “an extra year” if fully dubbed, a six-year production that grew Larian from under 50 people in 2014 to a 7-studio, 24-hour operation by 2023, and Early Access backlash when the game briefly dipped below 75% positive Steam reviews while Larian was “essentially doing games-as-service” just to sustain Act 1.
Larian then walked away from the D&D license entirely — no DLC, no Baldur's Gate 4 — to make something new. You only do that if you own the foundation.
My contrarian read: three ideas you won't find in press releases
Most coverage says “Larian succeeded despite using its own engine.” I think it succeeded because its engine is bad at the things Unreal is good at.
1. The cinematic tax is an anti-save-scum device.
1.5 million words of mo-cap sounds like bloat. I read it as emotional design. In older CRPGs, a failed Persuasion check is a line of red text — you instantly reload. In Baldur's Gate 3, failure gets actors, camera work, and music. Astarion sneers, Shadowheart intervenes, the narrator savors your humiliation.
When even failure is entertaining, players accept the dice. That single shift protects the entire D&D fantasy. A generic engine would have made cinematics cheaper per shot but harder to trigger systemically. Larian paid more per shot to make every shot triggerable.
2. Turn-based is accessibility arbitrage, not nostalgia.
Larian's 2020 pitch was that D&D is “easy to understand, difficult to master,” even playable in a car with kids and a mental d20. Real-time-with-pause, the Baldur's Gate 1/2 formula, hides math. Turn-based exposes it.
Exposed math streams well. Viewers can read action / bonus action / movement, predict outcomes, and backseat-drive. That turned Twitch into a tutorial system Larian didn't have to build. Owning the rules engine let them tune that legibility — highlight lines, surfaces, advantage flags — instead of fighting a generic action framework.
3. Shipping a neutered toolkit was deliberate scarcity.
Critics called the limited Patch 7 tools stingy compared to Neverwinter Nights' Aurora Toolset. I think it was smart product management. A full Divinity editor demands cinematic assets and voice acting to not look broken next to Larian's content — exactly why PC Gamer warned that new areas would struggle to match BG3's asset quality and VO expectations.
By shipping safe mods officially (items, classes, cosmetics for consoles) and letting Nexus unlock the dangerous level editor unofficially, Larian gets both: a curated console store without support liability, and a PC laboratory that may one day produce a total-conversion hit. It outsources R&D risk while keeping the mystique. That is dungeon-master thinking applied to business.
In short: Unreal sells you a faster horse. Divinity 4.0 is a slower horse that lets four friends argue about where to ride while the camera watches.
What indie devs can actually steal
You should not clone Divinity 4.0. You can clone its priorities:
- Pick the verb, then the renderer. If your fantasy is “talk your way out,” invest in dialogue state and save-forking before lighting. BG3's verbs are shove, throw, sneak, persuade — all cheap to render, expensive to simulate correctly in multiplayer.
- Make failure content, not a dead end. Even a two-frame reaction to a failed check beats red text. Players forgive RNG when the game performs the failure.
- Expose the math if you want streamers. Turn-based + visible dice + previewed hit chance turns every fight into teachable content. Hidden math kills clip-ability.
- Own the smallest tool you can't live without. Larian didn't write its own physics or audio middleware. It owned quest scripting, save branching, and cinematic triggering — the three places a licensed roadmap would have said “no.”
- Ship modding in layers. Official safe layer for everyone, unlocked power layer at-your-own-risk for experts. You get longevity without promising support you can't afford — exactly the lesson from Vincke's “we're not a tools company” warning.
Train the same muscles in your browser
You can't practice 100-hour CRPG planning in one sitting, but you can drill its sub-skills for free:
For branching consequences, try Fleeing The Complex. Its 5 endings — Ghost Inmate solo stealth, Convict Allies team-up, Presumed Dead brute force, International Rescue Operative government route, and The Betrayed Toppat twist — force the same question BG3 asks constantly: who do you trust, and what does that lock off? Play one run solo, one run with Ellie, and notice how early helper choice re-routes everything, just like picking Astarion vs. Shadowheart to lead a conversation.
For investigation logic, try Rooms Home Escape. Its loop — explore every corner, collect tools that stay useful longer than expected, connect a kitchen note to a living-room painting — is Baldur's Gate 3 exploration stripped of combat. If you take notes across rooms there, you'll do better in BG3's Goblin Camp or Moonrise Towers where clues span floors.
For action economy, try Master Chess. Chess has no dice, but its discipline — control center early, castle for king safety, think two moves ahead, don't trap your own pieces — maps directly to BG3 positioning. If you waste movement or cluster for a fireball in chess, you waste actions and eat AoE in Faerun the same way.
None of these replace BG3. They isolate one decision muscle so your next long rest in camp feels less overwhelming.
Scope, versions, and open questions
This analysis applies to Baldur's Gate 3 on PC and consoles as of final Patch 8 in April 2025, running Divinity Engine 4.0 with Vulkan and DX11 backends. Mod capabilities described reflect official Patch 7 tools plus community unlocks as of September 2024; console mod browsing remains curated and platform-approved.
Unresolved: Larian has confirmed its next game is codenamed Divinity, not Baldur's Gate 4, but has not detailed Divinity Engine 5 changes. Sales totals beyond “significant profits for Larian and Hasbro” vary by tracker and are intentionally not stated here. If you need exact current player counts or patch notes, check Larian's official channels before buying or modding.
Bottom line: the niche bet was not Vulkan, nor turn-based combat alone. It was keeping a boring, RPG-only toolchain for 13 years so that when Wizards of the Coast finally called, Larian could say yes to almost any player idea — and afford to animate the failure.
Top comments (0)