Cursor Vibe Jam 2026: The Complete Guide (Rules, Results, How to Build for It)

The Cursor Vibe Jam 2026 was the largest AI game jam to date, and it had a clear shape. Hosted by Pieter Levels and sponsored by Cursor, Bolt.new, Glif, and Tripo AI, it ran for one month, from April 1 to May 1, 2026, closing at 13:37 UTC. By the end, 945 games had been submitted and more than 242,000 people had played them, with the jam pulling around 12 million views on X. The prize pool was $40,000 in real cash, more than double the previous year.

This guide is the practical version: what the jam required, what those rules mean for how you build, and the exact workflow to ship an entry that fits the format. The format is the part worth understanding, because it is unusual and it decides everything, whether you are reviewing what happened or getting ahead of the next edition.

{/* IMAGE: A laptop on a desk at night showing a small fast-loading browser game on the left and a chat prompt on the right, with a sticky note on the screen bezel reading "WEB. NO LOGIN. 90% AI. ONE HOOK." 1200x630. */}

The two rules that decide your entry

Most jams give you a theme and a deadline. The Vibe Jam gives you a theme of sorts in its format, and the format is strict in two ways that quietly rule out a lot of tools and a lot of game ideas.

Rule one: at least 90 percent of the code must be written by AI. Not an honor system you can wave off, it is the premise of the jam. In practice your job is to direct, scope, and test while the AI writes and wires the game. The people who did well got fluent at describing a mechanic clearly, checking the result, and asking for the next change.

Rule two: the game must be playable on the web, with no login, no signup, free to play, and almost no load time. A player clicks a link and is in the game in a second or two, ideally on your own domain or subdomain. No installs, no account walls, no loading screen. This is the rule that strands people, because a surprising number of tools either cannot produce a web build at all or produce one so heavy it loads slowly. The official advice during the jam was blunt: keep the project light enough that it runs near instantly in a browser.

Smaller rules matter too. The game has to be new, made during the jam. One entry per person, so the instruction was to make one really good game rather than five rushed ones. A small JavaScript snippet from the jam goes on your page. Multiplayer was preferred but not required. And there was an optional touch: an exit portal players could use to hop from your game into the next one, turning the entries into a kind of webring.

Read those two big rules again, because every decision below follows from them. You are building a small, instantly playable web game where AI writes the code. That favors a specific kind of tool.

What the format does to your tool choice

People call a lot of different things an AI game maker, and the Vibe Jam format quietly sorts them into what works and what fails at submission time.

Browser text-to-game tools. You type a prompt and a small web game appears, already hosted, which clears the playable-in-a-browser rule with no export step. The catch is the same as in any jam: many cap AI generations on the free tier right when your demo is half built, some add a watermark, and most stall when you ask for the fourth or fifth interconnected mechanic. If your idea stays a single screen and the cap holds, this lane can carry you. Our honest look at the Rosebud-style approach covers where the wall is.

AI coding assistants. Cursor itself, and tools like it, generate code while you work inside an engine you already know. For an experienced developer this is a real boost with full control of the export. But it assumes you already know how to set up a web build, an input map, and a scene, because the assistant writes code faster, it does not build the game for you. Fine for a veteran, rough for a beginner. We go deeper on that split in Cursor plus Godot versus an AI native engine.

AI native engines. The AI is wired into the editor. You describe a mechanic, it writes the script, builds the scene, and runs the game, then you press play and ask for the next thing. This lane fits the Vibe Jam unusually well, because it satisfies the 90-percent-AI rule by design and the right engine exports the clean HTML5 web build the jam requires. Summer Engine is built this way and is compatible with Godot 4, so the file you submit is a real web build you host on your own domain. The honest caveat is the one the jam gave everyone: keep the project light so it loads fast.

No lane wins every time, but the web-export rule is the filter most people forget until the last night. Pick your tool by checking that filter first.

How to build a Vibe Jam entry, step by step

This workflow uses an AI native engine, the lane that satisfies both big rules at once, and it generalizes to any jam with the same web-playable, AI-written constraints.

1. Scope to one hook that reads in five seconds

The most important decision happens before you touch a tool. The jam rewards a small game a stranger understands instantly, because most players give an entry a few seconds before clicking away. Write one sentence describing the smallest version of your idea that is still fun to press a button in. Good Vibe Jam scope looks like:

• "You are a fish that has to eat smaller fish to grow, and everything bigger eats you."
• "A one button endless climber where tapping swaps which wall you stick to."
• "A tiny arena where you and the cursor are the same object and you have to escape yourself."

Each is buildable in a weekend, reads in five seconds, and runs in a browser. The version with an upgrade tree and three biomes stays in your head as the dream. What you ship is the sentence.

2. Start from a template that already runs, not a blank project

A blank project forces the AI to invent your player controller, camera, input, and physics before you have anything to feel, and each of those is an early bug you cannot afford. Start from a template nearest your sentence and reshape it. A platformer template already has a moving character and ground. A survivors-like template already has a moving character, swarms, and a survival timer, which is a strong web-jam base because the loop is satisfying with almost no content and stays light enough to load fast.

3. Build one mechanic per prompt, then press play

This is the rule that holds the whole build together. Make one change, run the game, feel it, then make the next change. So your first prompts look like:

1. "Reshape the player into a small fish and the background into open water." Press play. Confirm it still runs.
2. "Make the player grow when it touches a smaller fish, and end the game if it touches a bigger one." Press play. This is your hook, so spend your prompts here until it feels right.
3. "Spawn fish of random sizes from the edges and show a score for how big I get." Press play.

When a change breaks the build, and one will, you know exactly which prompt did it, and the AI fixes one small change far more reliably than a wall-of-text prompt that asked for six things at once. We recommend the same one-mechanic-at-a-time discipline for any deadline build in our hackathon workflow guide.

4. Keep it light so it loads near instantly

The no-loading-screens rule is a scoring factor, not a nicety, because a slow first load costs you players in the first second. Favor simple scenes, small textures, and a handful of sounds over a heavy download. If a big asset is not load-bearing for the hook, cut it. Most jam-winning games are tiny on disk.

5. Export the web build and host it

This is the step that the rules care about most. With Summer Engine you export an HTML5 web build the standard way, then host the generated files on your own domain or subdomain so the game opens with no login and no signup. Add the required jam snippet to the page, and if you want the portal touch, wire the exit portal so players can hop to the next game. Test the hosted link on a fresh browser with no account before you submit, because the most common late failure is a build that works locally but breaks when hosted.

6. Make the first ten seconds carry the whole thing

Players judge a jam entry fast, and so do tired judges scrolling 945 of them. Your opening has to communicate the hook with no instructions, so spend real time on the first thing a player sees and does. If they understand and smile in ten seconds, you are in the running. If they need a tutorial, the scope was too big.

The honest part about free versus paid

The Vibe Jam is free to enter, and you can build a complete entry for free. Summer Engine is free to download and use, including the web export the jam requires, with paid plans only for heavier AI usage and team features. For a single jam entry the free tier covers most people.

The honest catch is not specific to any one tool: some cap AI generations on the free tier and the cap runs out mid build, some add a watermark, and some lock the web export behind a paywall. Verify those three things on whatever you pick before you start, because in a jam they bite at the worst time. Summer Engine keeps the web export in the free tier on purpose, since a tool that cannot ship the file the jam asks for is not really free for a jam.

Getting ahead of the next one

The 2026 edition closed on May 1, with judging by Pieter Levels and a panel after the window shut. The Vibe Jam has run year over year, so the safe bet is another edition in 2027 around the same spring window, announced on X with about a month to build. There is no theme to study in advance, but there is a format, and the format is the hard part for most people.

The move that actually prepares you is to ship a small web game now, in the same constraints: AI writes the code, it loads instantly in a browser, no login, one strong hook. Do that once and the next time the jam drops you start building on day one instead of fighting your tools. If you want a starting point, pick a template and describe one mechanic, or read how the same deadline discipline works across any AI game jam. The games that win this format are small, fast, and clear, which is a skill you can practice on any free afternoon.

Sources: Vibe Jam 2026 official site, Pieter Levels jam announcement, Pieter Levels closing stats.