Baldur's Gate 3: The CRPG That Went Mainstream

In 1998 the original Baldur’s Gate arrived on five CD-ROMs and asked you to learn THAC0, a stat where lower numbers were better and nobody could explain why without drawing a diagram. BioWare’s Infinity Engine ran combat in real time and gave you a pause key to survive it, which meant every fight was a stop-motion argument between you and six people who kept walking into fireballs. It sold well for what it was. What it was, was a genre with a moat around it.

Twenty-five years later, Larian Studios released Baldur’s Gate 3 on PC on 3 August 2023 and put something like 875,000 people into it concurrently on Steam over the launch weekend. That is a Counter-Strike number attached to a game where you spend forty seconds deciding whether to use a bonus action. Something in the design broke the moat, and it is worth working out exactly what.

The turn is the load-bearing decision

Larian’s single biggest choice was made years before this game existed. The Ghent studio, founded by Swen Vincke in 1996, spent Divinity: Original Sin (2014) and Original Sin 2 (2017) building a turn-based tactical engine while the received wisdom said turn-based CRPGs were a museum piece and real-time with pause was the commercial format. When Larian took the Baldur’s Gate licence, the loudest early complaint from the old guard was that turns betrayed the series’ identity.

The turn is why the game is legible. Fifth Edition D&D — the ruleset Larian licensed from Wizards of the Coast — is built around an action, a bonus action, a reaction and a movement allowance per round. Rendered in real time with a pause key, that economy is invisible; you are managing an approximation of it through a queue. Rendered as a turn, it becomes a small puzzle with four visible pieces, and a person who has never rolled a d20 can read the whole board in a glance. Illegibility is what keeps people out of tactics games, and a turn is complexity you can see.

There’s a second-order effect. Because a turn is a discrete unit, a bad decision is attributable. You know precisely which choice cost you the fight, which makes losing instructive rather than mysterious. The Infinity Engine games lost you fights in a fog. This one hands you the postmortem.

The verb list is the actual invention

The ruleset is licensed. The engine is Larian’s. Where the two meet is where the game gets interesting, and the meeting point is a list of verbs.

Baldur’s Gate 3 lets you shove people as a bonus action. In tabletop 5e, shoving costs an attack — a real price. Larian discounted it to a bonus action, which sounds like a footnote and is in fact a design thesis. It means every ledge in the game is a weapon. It means a level designer placing a chasm is placing a damage source. Players worked this out in about a day and started building parties around gravity, and Larian, who had watched three years of early access telemetry since Act 1 went out on 6 October 2020, plainly knew they would.

The same logic runs through jumping (a bonus action that spends movement), throwing (anything you can lift, including a smaller party member), and surfaces. Grease, water, ice, fire, poison clouds — this is the Original Sin elemental sandbox smuggled into Faerûn, and it does the same job it did there: it converts the environment from scenery into inventory. A barrel is a spell you don’t have to prepare.

Why it works is a matter of authorship. A fireball is a solution the designer wrote for you. Stacking three crates, shoving a goblin priest off a bridge and setting the water on fire is a solution you wrote, and the game merely failed to prevent. Games are generous when they let you feel like the author of your own cheating. The real ancestor of that feeling is the immersive sim — Deus Ex and the Thief games teaching you that the intended path is a suggestion — and Larian’s contribution is proving the trick survives being turn-based and having a d20 in it.

The camera bought the audience

Here is the expensive part, and the part the genre could not previously afford. Every line in this game is performed. Larian’s own figures put the cinematic runtime north of 170 hours, on a script in the millions of words, with motion capture on conversations that a 2017 CRPG would have delivered as a portrait and a wall of text. The camera drops to shoulder height, the elf looks at you, and the genre’s oldest tax — reading — is quietly waived.

I want to be careful about what this achieves, because “voice acting” is a shallow reading of it. What the camera does is make refusal legible. When Lae’zel, the githyanki who joins you in the first hour, sneers at a decision, the sneer is a piece of feedback about the state of a system — approval, in this case — and the system is now readable without opening a menu. Larian used cinematics as a UI for the relationship model. That’s why the origin characters landed the way they did: Astarion, Shadowheart, Gale, Wyll, Karlach and Lae’zel are better surfaced than a good Planescape companion — a different achievement from being better written, and a harder one to fake.

The tabletop debt is real, too. The Dark Urge origin, the Inspiration points you earn for playing your background and spend to reroll a failed check, the way a failed persuasion opens a worse door instead of a load screen — this is a game that understands that the actual pleasure of D&D is the table’s reaction when the plan collapses. Failure as content is a hard sell in a genre trained on optimisation, and it’s the most quietly radical thing here.

Where it fights itself

The long rest economy is the design arguing with the fiction. Rests cost forty camp supplies and restore your spell slots, which is 5e’s core balancing act — resources across an adventuring day. In a video game with no dungeon master enforcing pace, the day is whatever you say it is. The optimal play is to rest constantly, which trivialises attrition; the honest play is to ration, which the game never asks you to do. Meanwhile most companion content is gated behind rests, so the game actively bribes you to break its own tension. Larian resolved this by shrugging, and I think it was the right shrug — but it’s a seam.

Inventory is the other one. Four characters, weight limits, and a loot table generous enough to bury you in gear you’ll never equip means real minutes of your life spent as a warehouse clerk. Pathfinder: Wrath of the Righteous has the same disease, as I’ve argued before; the maximalist CRPG keeps mistaking abundance for generosity.

And a note on the platforms, because it’s part of the story: the PS5 version is dated 6 September 2023, and the Xbox release has been publicly held up over split-screen co-op on the Series S — Vincke has talked about the memory problem openly rather than pretending it away. A game this size on a console this small is a real engineering fight, and Larian have been unusually honest about losing rounds of it.

The verdict

Baldur’s Gate 3 is the CRPG that stopped apologising. The genre spent two decades treating its own systems as an admission price and hiding them behind either nostalgia or automation. Larian’s answer was to make the systems the attraction — visible, physical, shovable — and then spend blockbuster money on the interface. The result is a game where the tactical layer and the character layer feed each other instead of taking turns, and where a first-time player and a 1998 veteran are looking at the same readable board.

It sprawls. Act 3’s city is denser than the game’s own pacing can support, and you can feel where the ambition outran the schedule. I’d take that trade every time over a game that never reaches.

If this is your entry point, the two places to go next are Disco Elysium, which proves how far you can push failure-as-content when you delete combat entirely, and Persona 5 Royal for a completely different answer to the same question: how do you make relationship systems feel like they cost something? Baldur’s Gate 3 is on PC now and PS5 shortly. It will still be worth playing in a decade, which is more than the 1998 original could claim by 2008.

Spoilers below

The Emperor is the structural masterstroke and the most divisive thing in the game, and both facts have the same cause. For most of the running time your protector is a voice in your head asking for trust it hasn’t earned, which means Larian built the central mystery out of the same substance as the tadpole mechanic — every illithid power you accept is an argument you’re losing. That the reveal makes some players feel manipulated is the design working. You were.

Astarion’s ascension is the sharpest fork Larian wrote, because it’s the only one where the “good” ending is unmistakably the smaller one. You can let a man take enormous power at a cost paid by thousands, and he will be happy, and he will be worse. Refusing it gives you a frightened person with two centuries of damage and no compensation. Most RPGs make the moral choice the powerful one. This one prices it correctly.

Karlach’s ending is the game’s honest admission that not everything is solvable. The engine gives you infinite verbs and none of them fix her, and Larian held that line through early access feedback that must have been brutal. The Netherbrain finale is the weakest hour — a big fight at the end of a game whose best fights are small ones — but the last conversation before it is the reason the whole enormous thing works.