Rated programming contests

Code, prediction files, models, and quizzes in one contest.

Solve code tasks, submit prediction files or CPU ONNX models, and take quizzes in rated rounds.

Practice the IOAI 2026 Individual Contest: six tasks, open until September 2027.

How it works

Train, compete, climb

rated

TrainPython, C, and C++ in your browser Lab

PracticeSolve past rounds in the problem archive

CompeteTimed rounds, one rating across formats

Several formats, one judge

Code, prediction files, CPU models, and quizzes share one contest and scoreboard.

Every task, one rating

All four task types count toward the same public ranking.

Judged like an olympiad

Pinned environments, a frozen scoreboard, and a locked results snapshot. Built for the IOAI.

For competitors

Climb one rating across every format.

Enter timed rounds, work through the archive afterwards, and watch your level move with every public contest.

Rated public rounds
Practice archive of past problems
Browser Lab, nothing to install

For organizers

Host a round your way.

Write Code, Prediction file, Model, and Quiz tasks in Contest Studio and run them on the open-source Judge.

Private and invite-only rounds
Locked task versions for reproducible results
Open-source, self-hostable Judge

Brunost Premium

Sign in

Use your Brunost account to enter contests, submit work, and review results in My Lab.

One workshop · three modes

Pick up where you left off.

Contests, practice problems, and your browser Lab, all scored by the same Judge.

Open contests Manage contests Open workspace

—

Account

Loading your account…

—

Activity

Loading recent activity…

—

Notifications

Search problemsTopic

Find a competitorContest tag

How rating works

One number, across every format

Your rating moves only when a rated contest finishes, and every format counts toward the same number: code, prediction files, models and quizzes.

It measures where you finish, not what you solve. Solving a problem everyone else also solved moves nothing. Finishing above contestants rated higher than you is what gains.

You move halfway to what the round argues for. We work out the place your rating predicted, compare it with where you actually came, and move you half of the way to the rating that result implies.

Nothing is created or destroyed. What you gain, the rest of the field gives up. Everyone starts at 1200.

See the arithmetic

Say you are rated 1200 and three contestants finish behind you: one rated 1000, one 1200, one 1400.

What your rating expected, against what happened
Opponent	Expected to beat them	You did	Surplus
1000	76%	yes	+0.24
1200	50%	yes	+0.50
1400	24%	yes	+0.76

Beating the 1400 is worth three times beating the 1000, because your rating said you probably would not. From those numbers we work out the place you were expected to take, compare it with the place you took, and find the rating that would have made your actual result ordinary. You move half of the way there — so the more wrong your rating was, the harder it moves, and a contestant who is plainly stronger than their number reaches their level in a few rounds rather than a few years.

A 400-point gap means the higher-rated contestant is expected to win about nine times in ten. Ratings run from 0 to 5000.

Levels
Level	Rating
Newcomer	under 1400
Pupil	1400
Specialist	1600
Expert	1800
Master	2100
Grandmaster	2400

Practice problems from the archive and unrated rounds never change your rating. You have to make at least one submission in a round to be rated by it — registering and not taking part costs nothing.

Loading profile…

Current contest

Select a contest to begin.

Problems stay in their own workspace so the contest brief remains easy to find and edit.

No contest selected

Build queue

Problems in this contest

Open a problem to edit its statement, tests, or evaluation.

Draft changes are saved per problem.

Project

Brunost workspace · browser kernel

Select a file

Program input (stdin)

TerminalBrowser WASM · project sandbox

Type help to see workspace commands.

Loading public submissions…

Control room

Platform administration

Keep the competition platform healthy, approve access, and operate the Judge fleet from one calm workspace.

Live control plane

Today at a glance

Everything in one place.

Monitor access, capacity, and communication without leaving the control room. Changes made here are recorded against your administrator account.

Last checked Waiting for data Automatic when this page opens

Service status

Operational Premium API and database

Worker capacity◇

— Loading Judge fleet

Pending accounts◎

— Waiting for approval

Identity links↗

— Explicit external mappings

Shortcuts

Common operations

↗

Operator note

Safe changes, visible history.

✓

Worker changes require a reason. Credential revocation needs an additional confirmation. Every result is stored with your account and timestamp.

Protected by administrator accessEmail link + authentication code or approved password account.

Accounts

Registration approvals

New accounts remain inactive until an administrator reviews them.

Requires review

Waiting for your decision

Approve trusted applicants or reject requests that need more information.

People

Users

Every account on the platform. Block stops sign-in and hides the person from rankings; remove erases the account.

—

Accounts

Search by handle, name or email.

User	Email	Roles	Status	Joined
Loading…

Oversight

All contests

Every contest anyone has created, including drafts, private rounds and archives.

—

Contests

Open a contest to manage it, archive one that should no longer run, or delete an archived one for good.

Contest	Creator	Status	Visibility	Problems	Participants	Created
Loading…

Execution plane

Worker fleet

Control capacity on the shared Judge without touching running submissions.

Judge controls

Recent worker operations

Who changed the fleet, what happened, and why.

Communication

Email centre

Keep transactional messages useful, consistent, and easy to review.

Resend delivery

Message templates

Choose a message type, edit both formats, preview it, and send a test to your administrator address.

Subject

Plain-text body

HTML body

Recent delivery attempts

Provider responses and delivery status, without storing message bodies.

Configuration

System connections

Review the explicit identity mappings that connect Brunost accounts to external services.

Protected

External identity links

Email matching never grants access on its own.

—

Active mappings managed by administrators.

Administrator session

Current access level and authentication policy.

✓

Sign-in policyOne-time link + code

Data boundaryPremium + Judge

Contest information

Details

Editable

Contest content

Problems

Each problem opens in its own editor. Nothing is expanded into this page.

Lifecycle

Publishing controls

Keep the contest in draft while you prepare the brief and tasks.

Post-contest

Leaderboard A

Freeze the eligible pre-deadline source attempts, run the separate evaluation, then release the immutable results snapshot.

1 · Basics

What are participants solving?

Required

TitleSlugLowercase letters, numbers, and dashes.

Task family

Code

Judge program output or a small CPU training task.

Choose one

PointsDifficulty · optionalSubmission limitIn a rated contest the default is 5 per 24 hours for tasks scored against hidden labels and 3 in total for quizzes; unrated contests and input/output problems get none. Attempts that crash or fail to run never count.Task content · Markdown

Your rendered task content will appear here.

Markdown · ready0 words · 0 characters

Write the complete participant-facing task here: explanation, constraints, examples, rules, and links.

2 · Evaluation

Choose how it is judged

Judge contract

Choose a scoring mode; family-specific fields appear below.

Reference languageJudge time limit · msMemory · MB

Reference solution

The Judge runs this trusted solution on each test input to produce the answer.

Algorithm code

Test dataset

Add inputs one by one or upload text files. Expected output is optional with a reference solution.

File typePlain text files: .txt, .in, or .input.

Input styleOne file is one complete stdin test. Write values with spaces or line breaks exactly as the contestant program reads them.

Output styleUse plain text stdout. A reference solution generates it; enter an answer key only when that option is selected.

No test inputs yet. Add the first one.

Scoring modeAll tests required gives full score only when every test passes. Per-test scoring gives equal credit for each passed test.

Simple authoring flow

Choose Algorithm, write the reference solution, add the test dataset, check the contract, and save. Premium stores the definition; the open-source Judge executes it.

Author check

Test the evaluation

Run a sample through the same Judge contract used for participant submissions.

Sandboxed

Sample submissionThis is executed only in the Judge sandbox and is never saved as a participant submission.

Code, prediction files, models, and quizzes in one contest.

Train, compete, climb

Several formats, one judge

Every task, one rating

Judged like an olympiad

Climb one rating across every format.

Host a round your way.

Sign in

Enter your code.

Pick up where you left off.

Contests

Public problems

Global ranking

One number, across every format

Public profile

Avatar

Password

Contest

Task

Manage your contests

Set the round up

Problem workspace

Select a contest to begin.

Problems in this contest

My Lab

Start with a project

Project

Submissions

Platform administration

Everything in one place.

Common operations

Safe changes, visible history.

Registration approvals

Waiting for your decision

Users

Accounts

All contests

Contests

Worker fleet

Recent worker operations

Email centre

Message templates

Recent delivery attempts

System connections

External identity links

Administrator session

Contest management

Details

Problems

Publishing controls

Leaderboard A

Define a problem

What are participants solving?

Code

Choose how it is judged

Reference solution

Test dataset

Quiz scoring

Optimization scoring

Test instances

ML evaluation recipe

Test the evaluation