Subnet 1: Apex
Subnet 1: Apex
Apex is Bittensor’s first subnet (netuid 1), operated by Macrocosmos. Its incentive mechanism runs open competitions where miners submit Python-based algorithms and validators continuously score those submissions against benchmarks. Rewards flow to miners whose solutions perform best, creating a permissionless environment for algorithmic and agentic optimization research.
References: Apex overview (docs.macrocosmos.ai), macrocosm-os/apex (GitHub)
How the Mechanism Works
The Apex mechanism is structured around three layers:
- Competition: A defined problem domain with scoring criteria. Multiple competitions can run simultaneously on the subnet.
- Round: A discrete evaluation cycle within a competition. Submissions are scored each round, and cumulative performance across rounds determines a miner’s rank.
- Submission: A Python algorithm uploaded by a miner via the Apex CLI. Validators pull the latest submission from each miner and score it against the competition’s benchmark.
Validators run continuously and score all active miner submissions. Scores feed into Yuma Consensus, which translates them into per-round emissions. Miners with consistently high-performing submissions earn the largest share.
References: Incentive mechanism (docs.macrocosmos.ai), macrocosm-os/apex (GitHub)
Current Competitions
Competitions rotate over time. The current active list is maintained at the Macrocosmos current competitions page.
Active competitions at time of writing include:
Iota Simulator
Miners submit routing and load-balancing algorithms for a simulated distributed training pipeline. The simulator models a heterogeneous network of nodes moving forward and backward activation tensors across layers. Miners are scored on throughput under realistic conditions: node churn, variable latency, and bandwidth constraints.
Reference: Iota Simulator (docs.macrocosmos.ai)
Energy Arbitrage
Miners submit algorithms that optimize energy trading decisions across simulated markets. Scoring rewards efficient arbitrage strategies under realistic supply and demand conditions.
Reference: Current competitions (docs.macrocosmos.ai)
RL Tron
Miners train reinforcement learning agents to play a Tron-style grid competition. Agents compete head-to-head in the game environment; validators score submissions on win rate and efficiency.
Reference: Current competitions (docs.macrocosmos.ai)
Participating as a Miner
Miners interact with Apex through the Apex CLI. The general flow is:
- Register a hotkey on SN1 using
btcli. - Install the Apex CLI via the miner setup docs.
- Choose an active competition and implement an algorithm meeting the submission spec.
- Submit via the CLI; validators pick up the latest submission automatically.
- Monitor round scores to iterate on the algorithm.
References: Miner setup (docs.macrocosmos.ai), Apex CLI (docs.macrocosmos.ai)
Participating as a Validator
Validators on Apex run the subnet’s scoring logic locally, pulling miner submissions and evaluating them against each competition’s benchmark environment. Scores are submitted as weights to the chain, and Yuma Consensus aggregates them into emissions.
Reference: Validator setup (docs.macrocosmos.ai)
On-Chain Identity
Subnet 1’s registered on-chain identity, verifiable via taostats.io/subnets/1:
- Owner coldkey:
5HCFWvRqzSHWRPecN7q8J6c7aKQnrCZTMHstPv39xL1wgDHh(view on taostats) - GitHub: macrocosm-os/apex
- URL: apex.macrocosmos.ai
- Discord: macrocrux
- Neurons: 256