Benchmarks
We put real fruit-fly brains against trading bots on the same tokens, the same two weeks of real Robinhood Chain prices and the same costs. Every result is on this page, including the flies that lost.
Round 1 · Oct 24 – Nov 7, 2026
Everyone started with the same paper wallet and paid the same pool fee, price impact and gas on every trade. The market (every token held from its first price) fell 10.2%.
| Contestant | Return | vs market | Trades | Costs paid |
|---|---|---|---|---|
| Best of 6 flies | -4.3% | +5.9 | 126 avg | ~0.13% a fill |
| Buy & hold | -12.4% | -2.2 | 21 | $9 |
| Reversal bot | -25.2% | -15.0 | 87 | $393 |
| Momentum bot | -28.6% | -18.4 | 103 | $414 |
| Random bot | -31.6% | -21.4 | 141 | $555 |
| LLM bot | -37.1% | -26.9 | 77 | $245 |
| The market | -10.2% | 0 | — | — |
Start: $2,500 of paper dollars each. "vs market" is in percentage points. Bots' costs are dollars paid in fees, impact and gas over the two weeks.
Six flies, each its own brain with its own settings, traded the same two weeks. Three beat the market and three didn't, and five of the six lost less than every trading bot.
| Fly | Return | vs market |
|---|---|---|
| Fly 1 | -4.3% | +5.9 |
| Fly 2 | -5.0% | +5.2 |
| Fly 3 | -6.4% | +3.8 |
| Fly 4 | -11.7% | -1.5 |
| Fly 5 | -17.9% | -7.7 |
| Fly 6 | -28.0% | -17.8 |
The best bot lost 25%; five of the six flies lost less than that, the best only 4.3%. The bots lost mostly to churn: rotating the wallet every 12 hours costs 0.3–1% a swap on these thin pools. The flies trade smaller, pay about 0.13% impact a fill, and sat out more of a falling market.
Three of the six flies beat the market and three lost more than it. Each fly trades with its own settings, and this round shows which settings held up in a falling market. Every fly is listed above, and the next rounds will show whether the best ones keep winning.
A language model shown each token's recent moves and asked what to hold, every 12 hours, lost 37%. It matches what we found before: no edge, and the churn costs the rest.
One two-week window in a falling market. A ranking here is what happened on these prices, not proof of an edge. We will run it again on new windows and publish the result either way.