12 open source tools compared. Sorted by stars. Scroll down for our analysis.
By Erik Loyd, SaaS CEO and former COO/CFO of an AWS Premier Partner.
| Tool | Stars | Velocity | Score |
|---|---|---|---|
Godot Multi-platform 2D and 3D game engine | 117.3k | - | 87 |
Bevy Data-driven game engine built in Rust | 48.3k | +123/wk | 88 |
| 26.8k | +168/wk | 86 | |
| 9.8k | +11/wk | 82 | |
mineflayer Create Minecraft bots with a powerful, stable, and high level JavaScript API. | 7.5k | +38/wk | 84 |
fheroes2 fheroes2 is a recreation of Heroes of Might and Magic II game engine | 3.5k | - | 67 |
forge An unofficial rules engine for the world's greatest card game. | 2.7k | +25/wk | 73 |
mtasa-blue Multi Theft Auto is a game engine that turns Grand Theft Auto: San Andreas into networked multiplayer. | 1.9k | - | 63 |
Tri-Inspector Free inspector attributes for Unity [Custom Editor, Custom Inspector, Inspector Attributes, Attribute Extensions] | 1.4k | +5/wk | 71 |
| 1.2k | - | 67 | |
lightyear A networking library to make multiplayer games for the Bevy game engine | 1.2k | +4/wk | 70 |
| 685 | - | 66 |
Stay ahead of the category
New tools and momentum shifts, every Wednesday.
Godot is a full game engine for 2D, 3D, mobile, desktop, web, and console games, completely free with no revenue share, no per-seat licensing, no strings attached. After Unity's pricing fiasco in 2023, Godot became the go-to for indie developers who don't want to worry about their engine vendor changing the rules. MIT license, C++. The editor runs on Windows, Mac, and Linux. Games can export to all those plus Android, iOS, and web. GDScript (Python-like language built for Godot) is the primary scripting language, with C# and C++ also supported. Built-in 2D and 3D physics, animation system, particle effects, tilemap editor, and shader language. Fully free. Zero cost. No revenue share. No per-seat fees. No 'free until you make money' clause. MIT license means you can sell your game, modify the engine, and never owe anyone anything. Solo indie developers: this is your engine. Period. Small studios: Godot handles teams well with its scene-based architecture (fewer merge conflicts than Unity). Medium studios: viable for 2D and stylized 3D. AAA-scale 3D: Unreal is still ahead. The catch: 3D capabilities are improving fast but still behind Unreal and Unity for photorealistic rendering. The asset marketplace is smaller. C# support exists but isn't as mature as GDScript. Console exports (Switch, PlayStation, Xbox) require third-party tools and console devkit access; Godot can't ship console export templates directly due to NDA requirements. For 2D games, there's no better option at any price.
Bevy is a game engine built in Rust around an entity component system. ECS means you build game objects by composing data components rather than inheriting from classes. It's a fundamentally different way to structure games, and it's fast. Apache 2.0, pure Rust. The feature list includes 2D and 3D rendering, audio, asset loading, UI, animation, and a scene system. Hot reloading lets you change assets without restarting. The plugin system means the community extends everything. Fully free. No paid tier, no royalties, no revenue share. Apache 2.0 means you ship commercial games without owing anyone anything. That alone differentiates it from Unity's licensing drama. Solo indie devs: the primary audience. Small teams comfortable with Rust: viable. Medium to large studios: not ready. Bevy is pre-1.0 and the API changes between releases. The catch: Bevy is pre-1.0 and the API breaks regularly. Migration between versions is real work. The Rust learning curve is steep if you're coming from Unity/C# or Godot/GDScript. There's no visual editor yet. Everything is code. And the ecosystem, while growing fast, is a fraction of Unity's or Godot's. You're betting on the future here.
GDevelop makes 2D and 3D games without writing code, using a visual event system that reads like plain English. It is game development for people who have ideas but do not want to learn C# or GDScript first. Drag behaviors onto objects, set up conditions and actions, export to web, mobile, or desktop. The editor itself is MIT licensed and free with no project limit locally. Everything you build locally stays free forever. What the subscription buys is the cloud layer: cloud project storage, cloud builds that spare you from installing the Android SDK or owning a Mac for iOS, leaderboards, player accounts, and multiplayer lobbies. You can export locally and pay nothing, which is a genuinely usable path if you are comfortable with the native toolchains. The free tier gives you 3 cloud projects, 1 cloud build per day, 1 leaderboard per game, and 4-player multiplayer, with GDevelop branding on the splash screen. Silver runs around $6 a month (€5.99) for 50 projects and 10 builds a day, Gold roughly $12 (€11.99) for 100 projects, 100 builds, and iOS, and Pro lifts the caps entirely. All three went up 20% in January 2026. Solo hobbyists live on free. Anyone shipping to app stores wants Gold for the iOS builds. The catch is that one build a day on the free tier is the real constraint, not the project count. Game development is iterate-test-iterate, and a daily build budget breaks that rhythm the moment you are debugging something device-specific. Also, the AI credits now bundled into each tier are a separate currency with their own weekly cap, so read what they actually cover before you factor them into the price.
Cocos Creator is a free game engine with a visual editor, scene graph, and cross-platform export to iOS, Android, web, and desktop. Cocos has been around since 2010 (originally Cocos2d-x) and is huge in Asia, particularly China. The engine handles rendering, physics, animation, UI, and audio. The editor is TypeScript-based, which is unusual for game engines but means web developers can jump in without learning C# or C++. It also exports to native platforms via C++. Completely free. The license situation is a bit unusual (custom license, not a standard OSI one), but there are no royalties, no revenue caps, and no paid tiers. Solo indie developers and small studios targeting mobile will find this compelling. The 2D workflow is particularly strong. For 3D, it's improving but still behind Unity and Godot in tooling maturity. The catch: English documentation is spotty. The community is predominantly Chinese-speaking. If you don't read Mandarin, you'll miss a lot of resources and forum help.
mineflayer turns a Minecraft account into a programmable client. Point it at a server, write JavaScript, and your bot can dig, build, fight, navigate terrain, and chat. MIT licensed, runs on Node, also callable from Python. There is nothing to host. Install the npm package, hand it a server URL plus credentials, and you have a bot. Pathfinding works out of the box. Where it gets hard is what you ask it to do. Coordinating multiple bots, surviving mob attacks, or building structures larger than a few blocks means writing real game logic. The library tracks underlying protocol versions, so new Minecraft releases usually work within an update or two. This is the standard for Minecraft automation. Hobbyists, modders, AI experimenters running LLM-powered bots, and researchers studying autonomous agents in 3D environments all use mineflayer because the alternatives are protocol libraries that hand you raw packets. Start here unless you have a reason not to. The catch: Minecraft is owned by Microsoft and the protocol changes with major updates. Expect occasional version pinning while maintainers catch up.
fheroes2 rebuilds the engine behind Heroes of Might and Magic II, the 1996 turn-based strategy classic, from scratch. You bring the original game files and it runs them with high-resolution graphics, better AI, bug fixes and a cleaned-up interface. GPL-2.0, free, funded by donations. Installation is the easy part. There are builds for Windows, macOS, Linux, Android, PlayStation Vita and Nintendo Switch, plus Flatpak, Homebrew and AUR packages. The installer pulls resources from a copy you already own, or downloads the free English demo if you don't. This is for people who want to play the game on modern hardware, and for developers curious how a long-running engine reimplementation stays alive. The team sends you to GOG or the Ubisoft Store for the full game, which keeps the project on the right side of copyright. Fans of the sequel should look at VCMI, which does the same job for Heroes III. The catch is scope. This is not a game engine you build new games on. It exists to run one specific game, and without the original data files you only get the demo.
Forge is a full rules engine for Magic: The Gathering that runs on Windows, Mac, Linux, and Android for free. This is not a fan-made toy: it enforces the actual game rules across a massive card pool, with draft, sealed, Commander, cube, and a single-player adventure mode in the spirit of the old Shandalar game. Setup is download, install Java 17 or later, and run. The Android build sideloads and pulls its own assets on first launch. It's GPL-3.0 and community-built, unaffiliated with Wizards of the Coast, which is exactly why it can cover formats the official products never will. Use it to practice limited formats, test decks against an AI that doesn't embarrass itself, or play Commander offline. MTG Arena stays the polished official experience; Forge covers everything else. The catch is the obvious one: it lives in Wizards of the Coast's legal shadow and always has. The interface is functional rather than pretty, and the AI still misplays complex board states.
Grand Theft Auto: San Andreas shipped in 2004 with no multiplayer at all. Multi Theft Auto (MTA) is the engine that makes that possible. It's a multiplayer mod framework that turns a 2004 single-player game into a platform for custom online experiences: racing, roleplay, zombie survival, whatever the server community builds. GPL v3, C++ core with Lua scripting for server-side game logic. The project has been active since the mid-2000s and still maintains an active player community. Servers run custom game modes written in Lua, and the modding API is extensive. You can alter nearly every game mechanic. Fully free. The mod client and server are free. You need a legitimate copy of GTA: San Andreas to play. Server hosting: run it on any machine or VPS. A small VPS ($5-10/mo) handles a decent player count. The Lua scripting API is well-documented and has a large library of community resources. The catch: this is a niche community project for a 20+ year old game. If you're not already in the GTA modding scene, this probably isn't for you. The codebase is old and complex. But for its community, MTA remains the most feature-rich multiplayer mod for San Andreas, beating the competing SA-MP (San Andreas Multiplayer) in scripting capabilities.
Tri Inspector makes Unity's inspector usable without paying for Odin. Add attributes to your fields and you get grouped and tabbed layouts, validation with inline error messages, conditional show and hide, buttons that call methods from the inspector, read-only fields, info boxes and reorderable table lists. It is the same category of quality-of-life work that makes designers stop asking engineers to change values for them. MIT licensed, distributed through OpenUPM. Nothing to run. Install the package from OpenUPM or a git URL and the attributes work. Version 2.0 is a preview release rebuilt on Unity's UI Toolkit, requiring Unity 6 or newer, with proper prefab override support so modified properties highlight, revert and apply correctly across prefab instances and variants. TableList rows are now drag-reorderable. If you need stability today, the 1.x branch is still there and still maintained. Free with no paid tier, which is the entire pitch given Odin Inspector sells for $55 per seat. Tri Inspector also integrates with Odin and Odin Validator rather than fighting them, so adopting it does not burn a bridge. Solo developers and small studios get most of Odin's daily value for nothing. The catch is that 2.0 is explicitly flagged as preview and may contain bugs, and it drops everything before Unity 6. Odin still wins on breadth, serialization of types Unity will not touch, and a support contract. Free covers the common cases, not all of them.
Castle Engine is a cross-platform game engine with a visual editor, physics, scene graph, and support for glTF, X3D, and other 3D formats. It compiles to Windows, Linux, macOS, Android, iOS, and web via WebAssembly. LGPL-with-static-linking-exception (effectively permissive for games), Pascal. The visual editor handles scene composition, and you write game logic in Object Pascal. Built-in support for 2D and 3D, skeletal animation, shadows, shaders, and a component-based architecture similar to Unity's. Fully free. No paid tier, no asset store fees, no royalties. The engine and all tools are open source. Solo Pascal enthusiasts: this is your engine. Everyone else: look at Godot. The catch is obvious: Pascal. The game development world runs on C#, C++, GDScript, and Rust. The Castle Engine community is small, tutorials are limited, and hiring Pascal game developers is nearly impossible. If you already know and love Pascal, this is a genuine option. The engine is well-built and actively maintained. If you don't know Pascal, there's no compelling reason to choose this over Godot, which offers a similar indie-friendly, fully-free experience with a vastly larger community.
Multiplayer netcode is the part of game development that quietly kills projects. Lightyear handles it for the Bevy engine: server-authoritative networking for Rust games with client prediction, interpolation, and input handling built in, free and open source. It's a library, not a service, so there's no infrastructure beyond the game server you were already going to run. Transport is pluggable: UDP, WebSocket, WebTransport (which makes browser builds work), and Steam sockets. Inputs are buffered per tick and protected against packet loss, and there are first-class integrations with leafwing-input-manager and the Avian physics engine. Bevy developers building anything competitive or physics-heavy should start here instead of rolling their own. The main alternative in the same ecosystem is bevy_replicon, which is simpler but leaves prediction and interpolation to you. The catch: it moves with Bevy. Every engine release means a migration, the docs are a work-in-progress book, and outside Bevy it's useless to you by design.
EzyFox Server provides the plumbing. It supports TCP, UDP, and WebSocket protocols, so it works for mobile games, desktop games, and browser-based multiplayer. It's free under Apache 2.0. You get a socket server framework, plugin system, cluster support, and client SDKs for multiple platforms. The architecture is designed for real-time gaming scenarios: low-latency messaging, room-based matchmaking, and user authentication. The catch: this is a niche project with a small community. Documentation is limited and mostly example-driven. If you're evaluating real-time game servers, Colyseus (JavaScript) has a much larger community and better docs. Photon (proprietary) is the industry standard for commercial games. EzyFox fits if you specifically want Java and self-hosted, but you'll be figuring out edge cases on your own. The homepage URL pointing to star-history.com instead of actual project docs is not a confidence builder.