A Hitman developer says GTA 6's NPCs are ten years ahead of the industry

Berkay Dursun, an AI programmer at IO Interactive who worked on the crowd technology in 007: First Light, has said Grand Theft Auto 6 is at least ten years ahead of the industry on non-player characters, based purely on the public footage.

The context is worth having, because it is not a press release. He was replying to a post that had compared Grand Theft Auto 6's crowds unfavourably with Hitman's. His answer was that the two are not the same kind of thing: in Rockstar's game every character is simulated individually, where a Hitman or Assassin's Creed crowd is mostly bodies running repeating movements with no schedule or behaviour underneath them.

Why the source matters

Developers praise each other's games constantly and most of it is professional courtesy. This one carries more weight because of who is saying it.

IO Interactive makes Hitman. The entire Hitman design rests on crowd simulation and NPC behaviour: routines that repeat believably, characters who notice things, disguise systems that only work because strangers evaluate what they are looking at, and levels that are essentially clockwork machines made of people.

No studio has spent more of the last twenty years thinking specifically about how background characters should behave. Dursun works on exactly that problem, and his own game's crowds are among the best in the business, so when he says a rival is a decade ahead he is conceding something on the one subject he is best placed to judge.

It is also worth noting what they are not saying. Nobody at IO is calling the graphics or the driving a decade ahead. The compliment is narrow and it is aimed at the hardest problem in the genre.

Why crowds are the hard problem

Rendering a hundred people is a solved problem. Making a hundred people appear to have somewhere to be is not.

The failure modes are familiar to anyone who has played an open world: pedestrians who walk into walls, who repeat the same four lines, who react to gunfire and then forget within seconds, who pop in behind you, or who exist in a bubble twenty metres wide and cease to exist outside it. Every one of those is a budget decision rather than an oversight, because simulation costs more than animation and the crowd is the largest thing on screen.

What Rockstar appears to be spending on instead is persistence and specificity: characters who are individually authored rather than assembled from parts, and who continue behaving when you stop looking.

What it lines up with

It sits directly alongside Rockstar's own account of the work: over 600,000 animations and a Los Angeles team of writers, directors and producers built solely to realise non-player characters.

That last detail is the striking one. A team of writers for background characters is not how most studios structure this. It implies the crowd is being treated as cast rather than as scenery.

It also matters mechanically rather than only atmospherically. The wanted system uses witness descriptions as evidence against you, and a witness system is only as good as the simulation of the people doing the witnessing. If a bystander is a prop, being seen means nothing. If a bystander is an agent who saw a face, a jacket and a car, being seen is the whole game.

The caveat

This is one developer's read of a trailer, offered in public and not on behalf of their employer. It is informed opinion about marketing footage, not a technical teardown, and trailers are assembled to look their best.

Crowd behaviour is also exactly the kind of thing that degrades under load. A scene built for a trailer runs at a density and a scripting level that a live open world at sixty simultaneous systems may not sustain, and Rockstar has not committed to a frame rate that would tell us how much headroom there is.

It is worth noting because of who said it. It is not evidence.