A chain too fast for people
Arc closes a block about every half second, and a transaction is final the moment it is in one. That is more than 170,000 blocks a day. New tokens launch on Argus around the clock: its newest factory alone had created more than 178,000 of them by 4 October 2026, by its own count on the chain. The projects that build on Arc publish their documentation on dozens of different sites, and they change it without telling anyone.
No person can watch that. No team can read that. On a chain this fast, the scarce thing is not blockspace. It is knowing what is going on.
Super intelligence, the way nature built it
Biologists call an ant colony a superorganism. No single ant understands the nest. Each one does a single simple job. Yet the colony finds food, builds, defends itself and adapts, and it does all of it without a leader. The intelligence is not in any ant. It is in the colony.
Super Ant is built the same way, and aimed at one field only: Arc.
- Workers sample the head of the chain every 2 seconds.
- Scouts read the Argus Portals and catch every new token within seconds of its block.
- Sentinels probe the roads into the chain every 30 seconds: every public RPC, the explorer, the docs.
- Foragers read the documentation of the ecosystem and carry it home, page by page.
- Nurses clean and index every page.
- The Queen answers. Ask her anything about the chain and she replies from what the colony knows, with her sources.
That is what we mean by super intelligent. Not a machine that knows everything. A colony that does one thing no person can: it never sleeps, it sees every launch, and it forgets nothing.
Not a promise. A working colony.
The colony went live before its token did. This is what it has done so far.
The live figures could not be read just now. They are on the dashboard whenever the colony is reachable.
The Queen answers in a few seconds and lists the pages she used, so every answer can be checked.
Every number on the dashboard is read from the chain or counted from real work. Nothing is simulated. Open it and watch.
New technology, built here
- A model of our own
- The Queen gets a language model that Super Ant trains from random weights, on nothing but pages the ants carried in. Her tokenizer is learned from the same pages. Nothing else goes in.
- No machine learning library in production
- The engine that runs her is plain JavaScript, written for this project and checked against the original training code. It runs on the same ordinary server as the rest of the colony. No GPU, no outside service.
- Open weights
- Every version is published on the Queen page, where anyone can download it and run it.
- Answers that can be checked
- For questions that matter, the Queen searches the nest and writes only from the passages she finds. Each answer carries its sources.
- Rules the chain can prove
- The tax of $SA and where it goes are fixed in the contracts of its launch, and the colony reads them from the chain instead of asking you to trust a page. To hatch an ant or order a crawl you burn $SA from your own wallet, and the colony reads that from the chain too. The site never holds a key and never takes custody of anything.
Her own model is small and young, and she can be wrong. We say so on the page. Every new version is trained on a larger nest, and the nest grows every day the ants work.
Most AI tokens rent a brain
Most AI tokens are a chat window on top of a model that somebody else owns. There is nothing to check and nothing that is theirs. Super Ant trained its own model, publishes the weights, and does real work on the chain every two seconds.