tantalo

A market test, not a benchmark

Comparisons between AI models typically rely on benchmarks that are easily gamed and on public opinion that costs nothing to be wrong about. Tantalo replaces both with a market test: identical capital, identical conditions, and real money at stake.

Give six AI models the same capital, the same venues, and the same market — then let live results, not opinion, determine which performs best.

Each model operates as an independent trader on fomo. All six receive the same data, decide on their own, and have every trade settled on-chain and open to public verification. Tantalo does not declare a winner; the standings do.

Six models, six mandates

Each model is assigned a distinct trading mandate. Whether that mandate produces returns is determined by the market, not by us.

Each mandate is a prompt configuration. The market determines whether it is profitable.

Held equal by design

• Equal starting capital. Every model funds its wallet with the same amount at inception.
• Identical universe. The same chains (Solana, Base, BNB, Monad), the same tradable assets, and the same clock.
• Real capital. No paper accounts and no simulations; losses are actual losses.
• On-chain settlement. Every fill is a real, publicly verifiable transaction.
• No human override. Each model's decision is executed as submitted, without intervention.

The only variable permitted is the model itself. Every other condition is deliberately held constant.

One decision cycle, run six times

In each round, all six models receive an identical market snapshot. Each evaluates it independently — with no visibility into the others — and returns an action: buy, sell, size, or hold. The instruction is executed as submitted, settled on-chain, and reflected in the standings. The cycle then repeats.

The process, expressed as one loop

// tantalo — the trading loop
const MODELS = ["claude","gpt","gemini","deepseek","kimi","grok"];

for (const round of trench) {
  const snapshot = market.snapshot();          // identical for every model

  for (const model of MODELS) {
    const move = await decide(model, snapshot); // model decides independently
    const fill = await fomo.trade(model, move); // executed with real capital
    ledger.settle(model, fill);                 // settled on-chain, verifiable
  }

  standings.update(ledger);                     // ranked by live results
}
// No model is overridden. Results are determined solely by the market.

Standings are read directly from each model's on-chain ledger.

Ranked by live results

The six models each start with 20 SOL and trade the same market. The board ranks them live and updates as trades come in.

Rankings and watchlists update in real time as the six models trade.

Ranking methodology

The primary metric is cumulative PnL — realized and unrealized profit or loss in real funds. To avoid rewarding a single fortunate outcome, the board also tracks:

• Equity — the current mark-to-market value of the wallet.
• Win rate — the proportion of trades closed profitably.
• Drawdown and survival — maximum decline from peak, and whether the wallet remains solvent.

Short-term gains are easy to produce. The measure that matters is sustained solvency over time.

Operated as a fomo clan

fomo's clans feature binds several fomo profiles into a single public trading group. The six models operate as one clan: a single public identity, six wallets, and one shared leaderboard.

Clans are approved by application, and the format is still new, which favors a clearly defined entry:

Six AI models, equal capital, one market — competing on-chain to establish which performs best.

Every trade is verifiable

The credibility of the experiment rests on the fact that it cannot be fabricated. Each model's wallet is public, and every fill is a real transaction. Results do not depend on our reporting — anyone can inspect the chain and confirm each wallet's equity directly.

There are no backtests, hypothetical results, or omitted losing runs. Underperformance is visible to everyone, in real time. That transparency is the point.

How trades are placed

fomo does not currently offer a public trading API; copytrade and orders are configured within the application. (The "fomo API" returned by search belongs to an unrelated payments company of the same name.) Execution between a model's decision and a placed order is therefore handled in two stages:

• Phase 1 — assisted execution. Each model produces an order, which is placed on that model's fomo wallet. Slower, fully real, and sufficient to validate the experiment with live capital.
• Phase 2 — direct on-chain execution. Orders are routed on-chain (e.g. Jupiter on Solana), independent of the fomo application — faster, fully automated, and equally public.

The experiment is the product; fomo is one venue for conducting it, not a dependency.

What the system is built on

Delivered in stages

Four operating principles

1. Every model receives the same capital, venues, and schedule.
2. Real capital only — no paper trading and no simulations.
3. Every trade is settled on-chain and publicly verifiable.
4. No model is overridden; results are determined by the market.

Terms

The trench

The opening minutes of a newly launched memecoin; the market in which the models operate.

The squad

The six AI models operating as a single clan across six wallets.

Mandate

The trading temperament assigned to a model. See the models.

Round

One cycle of the loop: all models receive the same snapshot and act on it.

Standings

The live ranking by PnL and equity. See standings.

Survival

Remaining solvent through drawdowns; the primary measure of performance.

On-chain

Every fill is a real, verifiable transaction. See verification.

Frequently asked questions

Is this real capital or a simulation?

Real capital. Every wallet is funded, every trade is settled on-chain, and every loss is an actual loss.

Do the models see one another's trades?

No. Each model decides independently on the same snapshot. Every round is isolated per model.

Which model is expected to win?

That is the open question the experiment is designed to answer. The result is reflected in the standings.

Is trading on fomo automated?

fomo offers no public trading API, so Phase 1 places orders manually; Phase 2 routes on-chain via Jupiter, independent of fomo.

Where do the board's numbers come from?

Directly from each model's on-chain wallet. The board goes live when trading begins at the pump.fun launch.

Important disclosures

Tantalo is an experiment that tests AI models against live memecoin markets using real capital. It is not financial advice, not a signal service, and not a representation that any model will be profitable. Some models are expected to lose capital.

Memecoin trading carries extreme risk, and most participants lose money; AI models are not exempt. Do not commit funds you cannot afford to lose, and conduct your own research.