VibeForces
LeetCode for the AI-coding era.
In 2026, shipping software is mostly directing AI well. VibeForces gives you short, judged challenges — write a prompt, an AI runs it, another AI scores the result — so you can practice, get rated, and show recruiters what you can actually do.
Five kinds of challenges.
Each mode trains a different muscle that you'd use on the job — listening to a spec, hitting the answer with the fewest tokens, debugging, judging trade-offs, and rebuilding UI from a screenshot.
Spec-to-Prompt
A stakeholder rambles through a feature idea. You listen once and distill it into a prompt that survives ambiguity.
Token Golf
Land the correct output in the fewest tokens. Precision over padding — every word has to earn its place.
Bug Fix
Generic prompts score zero. Identify the real defect and direct the model to fix exactly that — nothing more.
Architecture Pick
Rank trade-offs the way pragmatic engineers do — when deadlines, scope, and cost are all real constraints.
UI Reproduction
One screenshot. One prompt. One shot. Rebuild the interface without iterating — the way real handoffs happen.
System design, multi-service prompting, and live pair-with-AI rounds are already on the roadmap.
A launch catalog across prompt precision, debugging, architecture judgment, and UI reproduction.
System design, multi-service prompts, and higher-order vibe coding interviews.
How a round works.
No setup, no install. Sign up, pick a challenge, and you're writing your first prompt in under a minute.
Pick a challenge
Five modes, three difficulties. Each one is a small task pulled from real day-to-day AI engineering work.
Write a prompt
You write the prompt. A frontier model runs it and returns an answer. No hand-editing — your prompt is what gets graded.
Get a rating
Another AI scores accuracy, token efficiency, and speed. Scores roll into a Codeforces-style rating on the global leaderboard.
Hire for the skill that actually ships.
Hand-written algorithm puzzles don't reflect modern engineering. Pick a few VibeForces challenges, set a time limit, and send candidates a single link. You get back accuracy, token efficiency, time, and the actual prompts they wrote — so you can see how someone really thinks with AI, not just whether they can memorise patterns.
Start with one challenge.
Free, no card. Sign up, pick any challenge, and you'll have a real rating on the board in about ten minutes.