Our MOAT
Bittensor gives Swarm access to an always-on network of developers competing to build better autonomous-flight intelligence. Instead of relying on one internal team, we continuously attract new models, training strategies, compute, and technical approaches from around the world.
A global R&D network that compounds
Every benchmark cycle strengthens the system: better models raise the standard, harder missions expose new weaknesses, and the best contributors are rewarded for solving them.
Global AI talent, always on
A distributed network of developers trains and improves flight policies around the clock, giving Swarm a broader and faster R&D engine than a conventional centralized lab.
Incentives aligned with performance
Contributors are rewarded for measurable results, not activity. Models must outperform the current benchmark to earn rewards, creating continuous pressure to improve.
Model diversity creates robustness
Different teams, architectures, and training strategies compete against the same missions. This diversity helps uncover stronger solutions and reduces dependence on a single technical approach.
A compounding sim-to-real loop
Winning policies move from the open benchmark into Swarm's dedicated drone and robotics lab, where they are integrated with real hardware and tested in real-world conditions. What we learn feeds back into the benchmark, making the next generation of models stronger.
More contributors create better models. Better models strengthen the benchmark. Real-world testing makes the entire network more valuable.
How the network works
Swarm runs on Bittensor Subnet 124 — a perpetual, on-chain tournament where flight intelligence evolves through transparent competition.
Validators mint missions
Each epoch, validators generate fresh drone missions across six 3D worlds — new seeds, new starts, new goals.
Miners submit autopilots
A global network of developers trains flight policies and submits pre-trained autopilots to compete.
Rewards flow to the best
Every model is scored in the open. The top pilots hold the throne and earn the emission share.
→ An open arena where flight intelligence evolves through transparent competition.
Every model, proven in public
The flight models behind our lab are scored on SOTApilot — the same missions, the same worlds, refreshed every week for anyone to see.
5
Task families
6
3D worlds
1,000
Seeds per epoch
14 days
Benchmark epoch
Join our global network of developers
Swarm rewards developers for contributing smarter drone‑AI models and competing to be the best.
What is Bittensor?
Swarm is built on Bittensor, a global incentive network where nodes (miners) compete and collaborate to train better models.
What is a subnet?
A subnet is a specialized environment focused on one task. Swarm is Subnet 124 — autonomous drone flight, where miners contribute compute and training strategies for real aerial missions. How our subnet works
Why miners matter
In Swarm, miners produce intelligence — AI engineers from all over the world competing to deliver the best flight models. Every winning approach becomes part of the network, and rewards keep the strongest contributors improving.
Why it matters for drones
Real missions demand fast adaptation. A global training network learns in parallel at scale — producing robust, field‑ready AI faster than centralized labs. Smarter drones, safer missions, better autonomy.
Start mining in three steps
Get the repo
Clone the open-source miner and read the quickstart.
Run the simulator
Spin up the PyBullet environment and train your flight policy.
Submit & earn
Submit your model privately with one command — the best pilots earn TAO and get published as champions.
Beat the champion, take the throne
Each active family has a single champion. Submit a model that scores higher and you take its place — along with the emission share that comes with it.
