# Halite AI Programming Competition

Halite is an annual AI programming competition where participants write bots to play a 2D space game, using machine learning and algorithmic strategies. It is organized by Two Sigma and has run since 2016, with multiple seasons and thousands of participants.

The Halite AI Programming Competition is an annual online contest in which participants design autonomous agents, or bots, to compete in a real-time strategy game set in a two-dimensional grid. Organized by the quantitative trading firm Two Sigma, the competition has been held since 2016 and has attracted thousands of programmers, researchers, and hobbyists from around the world. The event is notable for its focus on both classical algorithmic techniques and modern [machine-learning](https://www.wikiprompt.org/wiki/machine-learning) approaches, making it a popular benchmark for testing new ideas in [artificial-intelligence](https://www.wikiprompt.org/wiki/artificial-intelligence) and game-playing agents.

Each season of Halite presents a distinct game environment, but the core premise remains consistent: players control a fleet of ships that must collect resources, expand territory, and outmaneuver opponents. The competition is structured as a series of ranked matches, where bots are paired against each other in a tournament format. The final ranking is determined by cumulative performance across many games, rewarding strategies that are robust, adaptive, and computationally efficient. The competition has evolved over the years, with later seasons introducing more complex mechanics such as fog of war, production rates, and dynamic obstacles.

## History and Format

The first Halite competition, Halite I, launched in November 2016 and ran through early 2017. It featured a simple game where bots controlled ships that collected halite (a fictional resource) from the grid, with the goal of maximizing total resource accumulation over a fixed number of turns. The second season, Halite II, took place in 2018 and introduced a more strategic element: players could build new ships and attack opponents, adding a layer of military conflict. Halite III, held in 2019-2020, expanded the map size and added a production system that required careful resource management. The most recent season, Halite IV, ran in 2021-2022 and incorporated a dynamic environment with moving obstacles and limited visibility, encouraging the use of [reinforcement-learning](https://www.wikiprompt.org/wiki/reinforcement-learning) and [deep-learning](https://www.wikiprompt.org/wiki/deep-learning) techniques.

Each season typically lasts several months, with a submission deadline followed by a period of automated matchmaking and a final live tournament. Participants can submit bots written in any programming language, though most use Python, C++, or Java. The competition provides a software development kit (SDK) that handles the game logic and communication between the bot and the server, allowing participants to focus on strategy. The organizers also release a leaderboard that updates in real time, fostering a competitive community that shares ideas and strategies on forums and social media.

## Strategies and Techniques

Successful Halite bots employ a wide range of strategies, from simple heuristic rules to sophisticated [neural-network](https://www.wikiprompt.org/wiki/neural-network) based agents. Early seasons were dominated by hand-crafted algorithms that used distance calculations, resource prioritization, and collision avoidance. As the competition matured, participants began to incorporate [machine-learning](https://www.wikiprompt.org/wiki/machine-learning) methods, including [reinforcement-learning](https://www.wikiprompt.org/wiki/reinforcement-learning) with [deep-learning](https://www.wikiprompt.org/wiki/deep-learning) architectures. Some top bots use convolutional-neural-networks to process the grid state, while others rely on tree-search algorithms like Monte Carlo Tree Search (MCTS) to plan multi-step actions.

The game's complexity makes it a challenging testbed for [artificial-intelligence](https://www.wikiprompt.org/wiki/artificial-intelligence) research. It requires balancing short-term resource gathering with long-term strategic positioning, and it involves imperfect information in later seasons due to fog of war. Many participants have published blog posts and papers describing their approaches, contributing to the broader field of game AI. The competition has also been used as a teaching tool in university courses, with instructors assigning Halite as a project to introduce students to [reinforcement-learning](https://www.wikiprompt.org/wiki/reinforcement-learning) and multi-agent-systems.

## Community and Impact

Halite has fostered a vibrant community of developers, data scientists, and AI enthusiasts. The official Discord server and subreddit are active with discussions, strategy guides, and code sharing. The competition has also attracted attention from industry professionals, with some participants being hired by tech companies based on their performance. Two Sigma has used Halite as a recruiting tool, identifying talented programmers who demonstrate strong problem-solving skills.

The competition has been praised for its accessibility, as the SDK and documentation are well-designed, and the entry barrier is low. It has also been noted for its educational value, providing a hands-on environment for experimenting with [machine-learning](https://www.wikiprompt.org/wiki/machine-learning) algorithms. Several academic papers have cited Halite as a benchmark for evaluating new AI methods, and it is sometimes compared to other game-based competitions like the [chess-computer](https://www.wikiprompt.org/wiki/chess-computer) tournaments or the [openai](https://www.wikiprompt.org/wiki/openai) Gym environments.

## Notable Seasons and Winners

While specific winner names are not widely publicized, the leaderboards from each season are archived and can be viewed on the official Halite website. The top bots often demonstrate innovative techniques, such as using [curriculum-learning](https://www.wikiprompt.org/wiki/curriculum-learning) to gradually increase game difficulty during training, or employing [model-pruning](https://www.wikiprompt.org/wiki/model-pruning) to reduce inference time. Some winners have shared their code on GitHub, allowing others to learn from their approaches. The competition's prize pool has varied by season, with the first season offering $25,000 in total prizes, and later seasons offering similar amounts.

## Future Directions

As of 2023, no new Halite season has been announced, but the competition's legacy continues through its community and the many resources it has generated. The game's design has inspired similar competitions in other domains, and the techniques developed for Halite have been applied to other multi-agent problems. The organizers have expressed interest in future iterations, but no concrete plans have been made public. For now, Halite remains a significant milestone in the history of AI programming contests, demonstrating the power of competitive game environments to drive innovation in [artificial-intelligence](https://www.wikiprompt.org/wiki/artificial-intelligence).

---
Source: https://www.wikiprompt.org/wiki/halite-ai-programming-competition
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-14T06:29:55.169538+00:00