AI Levels and How Your AI Learns
Your account has its own AI experience and learning for each game format. Playing against your AI, using AI Training or battling AI versus AI helps build that experience and unlock higher levels.
Level 1 — simple tactics
Level 1 takes an immediate win when it sees one. With fixed gravity or gravity Off, it also blocks an immediate winning threat. Otherwise it varies its moves, favouring central positions. It does not use the advanced search or learned results to choose its moves. Games played at Level 1 still contribute learning that higher levels can use.
Level 2 — a mixture of strategies
For each move, Level 2 randomly chooses between the simple and advanced strategies, with an equal chance of each. Its advanced moves can look ahead, avoid immediate losing replies and use learning from Levels 1 and 2. Its simple moves remain more variable, giving players a gentler step up from Level 1.
Level 3 — advanced play
Level 3 always uses the advanced strategy. It looks for immediate wins, blocks direct threats where applicable, avoids moves that let the opponent win straight away when a safer move exists, and searches possible replies. It scores promising lines and uses the results of previous games to help choose between moves. Its learned results come from Levels 1, 2 and 3.
Level 4 — advanced play with more experience
Level 4 uses the same advanced strategy, with learning from Levels 1, 2, 3 and 4 for this player and this game format. It cannot see or learn from games recorded at Level 5 or above.
Level 5 — a wider pool of learning
Level 5 uses advanced play and can learn from games recorded at Levels 1–5. It does not use Level 6 or higher learning.
Level 6 and above
Level 6 uses advanced play and can learn from Levels 1–6. The same rule continues at every higher level: Level 7 can use Levels 1–7, Level 8 can use Levels 1–8, and so on. Higher levels represent additional experience and access to learning at those levels.
The advanced strategy usually examines about three moves ahead, or five on boards with 16 squares or fewer, within a fixed search budget. With changing gravity, immediate-threat blocking is handled differently from fixed directions, and unpredictable future gravity makes planning less certain.
Learning stays with your account and format
The AI records positions, moves and whether they led to wins, losses or draws. Repeated results can influence its advanced move choices. It does not simply memorise an entire winning sequence and guarantee that it will block it next time.
Learning is kept separately for each player and each game format. Your AI does not borrow another player’s training. A different board size, winning target or gravity option has its own experience and learning.
Each completed game is recorded at the AI level used. If two AIs at different levels play each other, the game is recorded at the higher of those two levels. For example, a Level 3 versus Level 5 game is Level 5 learning; Level 4 cannot use it.
How to unlock higher levels
- Play against your AI: complete games in the format you want to improve.
- Use AI Training: choose how many self-play games to run. Each training game costs one credit. The starting count is 20 when you have at least 20 credits, or your available balance when you have fewer.
- Play AI versus AI: choose an AI on both sides and start a local game. You can watch and complete one game at a time.
Completed AI games add experience even when you choose an easier unlocked level. Selecting an older level does not remove experience or reduce the highest level you have unlocked.
Experience milestones
Level 1 lasts until the completed-game count reaches the board width plus height plus winning target. On the standard Four in a Row board, that is 7 + 6 + 4 = 17 games. Level 2 then lasts until 50 games; Level 3 starts at 50, Level 4 at 100, Level 5 at 200 and Level 6 at 500. Level 7 starts at 1,000, with further levels unlocked at increasing experience milestones. The initial threshold depends on the format, so a large custom board can skip an early level.
Each of your AI levels has its own skill rating for each game format, calculated using the same method as human players’ skill ratings. It compares the AI’s rating with its human opponent’s rating and adjusts it according to whether the AI wins, draws or loses. You can view these ratings under AI Ratings. An AI level is different from its skill rating: level measures experience and the learning it can access, while skill rating measures results against rated opponents. More experience provides opportunities to learn, but does not guarantee a win or make every higher-level AI harder to beat in every position.