A memecoin that funds open-source frontier AI.
Hero Run is a real AI product wrapped around a token. People pay $HERO to run any model; the token's trading fees, product margin, and the data every run creates all flow toward one goal: training open-source AI models anyone can run.
Frontier AI is consolidating behind a few labs.
cost of a single frontier training run today (≈2× with RL).
labs on earth that can raise the $100M+ to do this with low risk.
open-source AI falls behind because of funding, not ability.
A flywheel that spins on something real.
Every memecoin claims a flywheel: buy, price rises, attention, more buys. It spins on nothing, so eventually it stops. The True Flywheel puts a load on the wheel. Each turn converts trading into open-source AI: a trained model and a dataset that exist whether the price is up or down. That output is what pulls in the next turn.
Pay $HERO to run any of 338 models.
Each spend and buy is a trade; every trade pays a 1.2% swap fee.
0.665% of every trade lands in the treasury, on-chain.
It pays to train open-source models anyone can run.
The volume that worthless memecoins already generate (hundreds of millions in a day) becomes the funding source for open AI. That is the difference: this token's volume does something, which is the reason it can hold value the empty ones can't.
How fast volume funds a $10M run.
Live market data from DexScreener (Base). Scenarios illustrate the model; they are not forecasts. See the full economics.
If $HERO traded like a coin you know.
Not a forecast, a yardstick for how hard the flywheel can spin. Take each coin's real 24-hour volume, run it through the same 0.665% fee, and this is what the treasury would raise for open-source training. At Dogecoin's daily volume, $HERO funds a $10M frontier run every few days.
24h volumes loading. A hypothetical of the fee model. Reaching these volumes would take an enormous, sustained market.
Usage gives three things.
Real revenue after conversion and hosting, on every run.
0.665% of every trade to the treasury. Scales with speculation volume.
Every run logs a prompt→output pair (opt-in), training material for the open models.
Hero Run routes across the whole inference market.
$HERO is not tied to one vendor. Under the hood Hero Run is a routing layer over many LLM gateways: today OpenRouter, Groq, and Cerebras, with more slotting in behind the same interface. When several gateways serve the same model, we route to the cheapest and fail over to the rest. Cheaper routing means a wider margin, and that margin funds open-source AI. The routing policy feeds the mission.
Text, image, video, and audio routed across ten gateways. One token, one endpoint.
The same model on three gateways collapses to one entry, priced at the lowest and backed by the others.
Payments verified on Base before any model runs. Treasury and fees are public and auditable.
No signup. Every newest model. One endpoint.
Most tools make you make an account, then a second account per model provider, then juggle API keys. Hero Run collapses that into one wallet and one token.
No account, no API keys, no per-provider signups. Connect a Base wallet and run any model.
375+ across ten gateways, the latest releases in one catalog. New models slot in behind the same interface.
Any AI agent plugs into a single MCP server and runs them all, paying $HERO per call from its own wallet.
And it keeps growing. New gateways slot in behind the same interface, and new models — text, image, video, audio — appear as providers ship them. The catalog is never finished.
An agent gateway that gets cheaper as it grows.
Hero Run is also one API for agents: an MCP gateway where any agent runs 375+ models and pays $HERO per call from its own wallet. Every call funds open-source training, and as those models get cheaper to run, the same work costs the agent less $HERO over time.
One gateway, 375+ models. The wallet is the account, no keys, no signup.
Each call is priced off the live token price and the model's real cost. As both improve, a run costs fewer $HERO.
Fees train cheaper, better open-source models over time.
The same $HERO runs more capability as the models it funded improve.
More agent usage → more funding → better, cheaper open models → each run costs less $HERO → more reason to route agents through the gateway. The gateway funds the thing that makes its own usage cheaper.
Your $HERO buys more compute over time.
There is no token burn. The deflation is in the price of compute, not the supply of the token. Every run is priced live: the $HERO you pay equals the model's real USD cost divided by the live $HERO price, plus a margin. Two things push that number down as the network matures.
The treasury funds open models that run for a fraction of today's cost. As inference gets cheaper, the USD cost of a run falls, so it takes less $HERO to pay for it.
As usage and demand grow, a higher $HERO price means the same USD cost is covered by fewer $HERO. Both levers point the same way.
Put together, the same capability costs fewer and fewer $HERO as the network matures. That is the flywheel: fund cheaper open models, each run costs less $HERO, the token does more work, which pulls in more usage, which funds more models.
The early-adopter case. Today a run costs a lot of $HERO and the token is cheap. As the open models we fund get cheaper to run and the token appreciates, that same run costs a fraction of the $HERO. Early holders buy compute at the network's most expensive, least efficient moment, and watch their $HERO stretch further with every model the treasury funds.
$HERO is a volatile utility token; price can fall as well as rise. This describes the pricing mechanism, not a promise about returns. Nothing here is financial advice.
We don't build a lab. We fund the ones already doing it.
The treasury deploys fees as compute and research funding to the open, decentralized training networks already shipping open models, so the ecosystem puts out as many as possible. Buying GPU time on decentralized markets, backing open runs, and funding the environments and datasets they need.
Distributed-RL open models (INTELLECT-2 32B, INTELLECT-3 100B+ MoE), a global compute marketplace across 50+ datacenters, and open frameworks like Prime-RL and Verifiers. Buying their compute and backing peers like them is how fees become open models. Named as an example, not a partnership.
The full plan, from the first funded model to a frontier run, is on the roadmap.
Open AI shouldn't depend on a handful of labs.
The frontier is getting more expensive and more closed. A run that cost under $20M a year ago now has a ceiling ten times higher, and the labs that can afford it are counted in dozens. Open-source keeps the field honest, but it runs on grants and goodwill that dry up.
We think the internet is good at exactly one thing at scale: moving attention and volume through tokens. Hero Run points that energy at a real cost. If a community can fund open models the way it funds a joke, and keep funding them, the joke was worth telling. That is what we are testing.
See it work.
Run a model, watch the payment settle on-chain, read the live economics. Everything here is real.