Anyone Can Join Fun Competitions and Earn as Miner on Apex (SN1)

Subnet 1 (SN1) Apex is part of the Macrocosmos ecosystem, focused on game-theoretic AI and open-ended intelligence. Instead of a single mining task, SN1 runs multiple live competitions where miners compete by submitting algorithms or models.

Each competition has its own incentive pool, meaning miners can specialize in different problem areas while emissions are distributed across parallel tracks.

Right now, SN1’s active competitions include:

• Matrix Compression (Lossless)
• Matrix Compression v2 (Lossy)
• Reinforcement Learning: Battleship

These competitions are built for practical AI outcomes, including improving distributed training performance for other networks like Subnet 9 (IOTA).

1) Matrix Compression (Lossless)

This challenge focuses on lossless compression of neural network activations, including forward and backward pass tensors. The goal is simple: reduce data size while keeping outputs perfectly identical.

The long-term goal is real-world utility, with winning approaches potentially being integrated into Subnet 9 (IOTA) to reduce memory and communication overhead.

Key Rules

• Optimize for compression ratio and speed
• Must be fully lossless
• Any loss leads to a score of 0
• Must beat the current top score by at least 1%
• Round duration: 2 days
• Submission cap: 4 submissions per 24 hours (network-wide across competitions)
• Max file size: 20KB
• Code stays hidden for 24 hours after submission

Incentive Mechanism

• 90% burn rate
• The top scorer earns the remaining 10%
• Rewards decay linearly over 10 days
• A new top submission resets the reward cycle

How to Join

• Download dataset manifest: manifest.csv
• Access matrices via R2 bucket URLs
• Use baseline GitHub code
• Install allowed packages via requirements.txt
• Submit through the Apex CLI
• Track rankings via the dashboard

2) Matrix Compression v2 (Lossy)

This is the higher-risk, higher-upside version of the compression challenge.

Here, minor loss is allowed, as long as the decompressed matrix stays close enough to the original to remain usable for training. It focuses on bfloat16 matrices and is also aimed at supporting improvements for Subnet 9 (IOTA).

Key Rules

• Optimize compression ratio + similarity score
• Similarity measured via norm × cosine similarity
• Must hit a minimum similarity threshold or score becomes 0
• Must beat baseline or top score by 1%
• Round duration: 2 days
• Max file size: 20KB
• Submission cap: 4 per 24 hours
• Code hidden for 24 hours

Incentive Mechanism

• 90% burn rate
• Winner earns the remaining 10%
• Decays over 10 days, resets when surpassed

How to Join

Same pipeline as the lossless version:

• manifest.csv + R2 bucket dataset
• baseline GitHub repo
• requirements.txt packages only
• submit via Apex CLI
• monitor via dashboard

3) Reinforcement Learning: Battleship

This is SN1’s first major reinforcement learning competition, built around the classic Battleship game.

Miners train RL models to locate and sink hidden ships using trial-and-error learning. The goal is to win consistently while using the fewest turns possible.

It’s also a signal that SN1 is leaning into renewed interest in RL as breakthroughs continue to push the space forward.

Key Rules

• Model must be submitted as TorchScript (.pt)
• Must load using torch.jit.load()
• Max file size: 100MB
• Multiple games per match
• No repeated shots allowed

Scoring

• Win: 1000 points
• Speed bonus: (100 – turns_to_win) × 0.1
• Final score is the average across games

Submission + Incentives

• Code is revealed 1 day after evaluation
• Submission cap: 4 per 24 hours
• Incentives follow SN1’s burn + annealing mechanism
• Winner earns the competition emissions (adjusted for burns)

How to Join

• Train locally using the train folder in the GitHub repo
• Install dependencies via requirements.txt
• Submit your .pt file using the Apex CLI
• Watch match replays after evaluation
• Track results via the dashboard

Getting Started on SN1 Apex

To participate in any of these competitions, you need to be properly set up as an SN1 miner. All submissions are handled through the Apex CLI, and rate limits apply across the network.

Competitions evolve quickly, so the dashboard is the best place to stay updated on current standings and rule changes.