Box3D: Erin Catto Releases an Open Source 3D Physics Engine

Erin Catto, the creator of Box2D, just released Box3D - an open source 3D physics engine that extends Box2D's design philosophy into the third dimension. If you've built browser games in the 2010s, you probably used Box2D without knowing it. This release gives game developers a new option in the surprisingly thin field of open source 3D physics engines.

What's In Box3D

The engine is written in C17 with a clean C API. It includes:

• Triangle mesh and height-field collision detection
• Baked compound collision systems
• Sub-stepping solver with continuous collision detection
• Graph coloring for handling large simulation islands
• Wide SIMD contact solver
• Multi-threading hooks with optional internal scheduler
• Large-world support using double-precision positioning
• Cross-platform determinism with recording/replay capabilities

That last point matters more than it sounds. Cross-platform determinism means the same simulation inputs produce identical outputs regardless of which platform runs the simulation. This enables replay systems, networked physics synchronization, and test reproducibility - features that are notoriously difficult to achieve with floating-point physics.

The codebase blends Box2D algorithms with elements from Rubikon-Lite, Valve's physics engine from Half-Life: Alyx. According to the announcement, Dirk Gregorius at Valve developed optimizations in a new engine called Ragnarok that influenced Box3D's architecture.

The Open Source 3D Physics Landscape

Before Box3D, the open source 3D physics space was remarkably sparse. The HN discussion highlights the history:

"The ancient forefathers are ODE, Bullet and Newton Dynamics (all first released in the early 2000s), then nothing(?) for nearly two decades until Jolt in 2021 and now Box3D."

Jolt (used in Horizon games) arrived in 2021 and quickly became a go-to choice. PhysX went open source in 2018 but carries NVIDIA baggage. Bullet remains widely used but shows its age. Rapier brought Rust to the party but has different design tradeoffs.

Box3D enters this space with specific advantages: it's from a proven physics engine author, it has Valve production heritage, and it prioritizes determinism from the start.

What HN Is Saying

The HN thread (365+ points, 80+ comments) is notably positive, with several game developers chiming in.

On Box2D nostalgia:

"Box2D was a foundation for a lot of interesting physics oriented indie games in my day. I wonder if the landscape is empty enough for a resurgence."

The thread name-drops IncrediBots, Angry Birds, and dozens of Flash-era physics games that used Box2D under the hood.

On the determinism features:

"I was looking for the same thing. There is a replay mechanism, so it seems to be deterministic. But with floating point physics, not across platforms. Though -ffast-math is unsupported according to the documentation, so maybe it is intended to be deterministic across platforms?"

A commenter found the answer in the documentation: "Box3D is designed to be deterministic across thread counts and platforms." This is a significant engineering achievement for a physics engine.

On Valve's involvement:

"On the Valve side, Rubikon continues to evolve and Dirk has developed optimizations (similar to those in Box3D) in a new engine called Ragnarok. Look for that in future Valve games."

This triggered the predictable Half-Life 3 jokes, but also genuine curiosity about Valve's upcoming physics-heavy projects. One commenter mentioned a Valve game codenamed "HLX" that apparently uses extensive physics features.

From Glenn Fiedler (gafferongames):

"Yeah this library is great. Use it!!!"

Glenn Fiedler is one of the most respected voices in game networking and physics. He's using Box3D in a 1000-player space game, which is a meaningful endorsement for networked physics use cases.

Current Users

Beyond Fiedler's space game, Box3D already powers:

• The Legend of California (Kintsugiyama Studios) - the primary development testbed
• s&box (Facepunch Studios) - which notably ripped out Source 2's physics for Box3D
• Esoterica engine

The s&box move is particularly interesting. Facepunch explicitly chose Box3D over Valve's native Source 2 physics, suggesting the open source option has real production advantages.

Getting Started

Box3D follows the same build pattern as modern Box2D:

git clone https://github.com/erincatto/box3d.git
cd box3d
cmake -B build
cmake --build build

The API is documented in Doxygen headers, and sample code is included. The current release is alpha software targeting a v1.0, with planned improvements for character movement, ghost collision mitigation, and joint solver refinements.

My Take

The open source 3D physics space needed this. Not because existing options are bad - Jolt is excellent, Rapier is great for Rust projects, PhysX is comprehensive - but because more good options push the whole field forward.

Box3D brings specific strengths: deterministic cross-platform physics (hard to find), production heritage from both Box2D and Valve, and an author who has been thinking about physics simulation for decades. The clean C API also matters - it's bindable to essentially any language ecosystem.

For indie game developers evaluating physics engines in 2026, the realistic choices are now Jolt, Rapier (if you're in Rust), PhysX (if you want the full commercial package), or Box3D. Each has different tradeoffs. Box3D's bet is on simplicity, determinism, and the accumulated wisdom of someone who's been optimizing 2D physics simulations since 2006.

If you're building something that needs networked physics or replay systems, the determinism guarantees make Box3D worth evaluating first. If you just need physics that works, any of the options will serve you well.

Sources

• Box3D Announcement
• Box3D GitHub
• Box3D Documentation
• HN Discussion