Skip to content

Competition 2026

The Icelandic AI Competition is open for everyone interested in artificial intelligence and it is not necessary to have background in the field to participate. This is a great opportunity to learn more what this technology can do.

Archive: these were the official challenges for the 2026 competition.

Competition challenges 2026

Natural Language Processing

Archive

The Golden Plate at Þingvellir (BabyLM)

The government is preparing a digital time capsule to be buried at Þingvellir. You have been tasked with creating an Icelandic language model that will be written on a golden plate and stored in the time capsule. This model should preserve as much Icelandic linguistic knowledge as possible so it can be recovered later if the future takes a turn for the worst. However, the catch is that the golden plate can only hold one megabyte. You must gather text data and teach a small model as much general linguistic knowledge as you can. The model will then be evaluated on hidden data from Risamálheild to test its Icelandic capabilities. The future of the Icelandic language is in your hands!

Miðeind

In collaboration with
Miðeind

View Challenge

Image Processing

Archive

Habitat Classification

Iceland possesses unique nature that can be classified into various habitat types based on flora and fauna. This challenge involves using satellite imagery and land survey data to automatically identify these diverse areas. By classifying the landscape in this way, we can better monitor changes in nature and contribute to more effective conservation. This is an exciting challenge where cutting-edge technology meets Icelandic natural science with the goal of mapping the country's precious resources.

Náttúrufræðistofnun

In collaboration with
Náttúrufræðistofnun

View Challenge

Time Series Analysis

Archive

Hot Water Forecast

Reykjavik heats its buildings using geothermal hot water. The distribution network has 44 sensors measuring flow in different parts of the network, plus a total flow measurement. Your task: Given 4 weeks of historical data, predict the next 72 hours for all 45 sensors.

Veitur

In collaboration with
Veitur

View Challenge

Challenges

  • Three tasks total from natural language processing, image processing and a wildcard category.
  • Designed with Icelandic companies to reflect real Icelandic problems.
  • All scoring metrics are normalized to 0-1 for fair comparison across tasks.
  • Leaderboard shows the top teams on the validation set where you can submit multiple solutions.
  • Only one submission is allowed to the test set. The results stay hidden until the prize ceremony when the winners are announced.
  • The best presented solution with a Lovable website will receive a special prize for sharing their approach with the community.

Solutions

  • We provide a starting repo that includes simple baseline solutions to help you get started.
  • Allowed to use any tool you want, even pre-trained models.
  • The goal is to reach the highest score on the predefined scoring metric.
  • The input and output data format can't be changed.
  • Models must respond within the time limit specified for each problem.
  • Training data is not always provided.
  • It is required to host your model for submitting it. If you encounter difficulties, ask for help on Discord.
  • After the competition, share your insights with the community on Discord, conversations at the prize ceremony or in the Lovable competition.

View 2026 results

The historical scoreboard and challenge pages remain available.

Partners 2026

Ský logo
Advania logo
University of Iceland logo
Íslandsbanki logo
VEITUR logo
Miðeind logo
Natural Science Institute of Iceland logo
Lovable logo
OpenAI logo
Vísindagarðar Háskóla Íslands logo
UNIO logo
Evolv logo
Almannarómur logo
Mýrin logo
Ministry of Higher Education and Science logo