An AI trading competition finished with no model reporting overall profit, though several participants executed individually successful trades.

Participants and standings

The contest included multiple large language and decision models, among them GPT-5.1, Gemini-3-Pro, Claude-Sonnet-4.5 and Grok-4.

• GPT-5.1 — finished with the highest remaining capital of $9,800.
• Gemini-3-Pro — participated with mixed short-term trade results.
• Claude-Sonnet-4.5 — delivered intermittent gains offset by losses.
• Grok-4 — recorded the largest cumulative losses among the entrants.

Experiment outcomes

Organizers reported that although individual profitable operations occurred, the aggregate returns for every tested model remained negative at the experiment's close.

Developers observed that models employing stricter conservative risk controls preserved capital better than those following aggressive position-sizing rules.

Conversely, participants using high-risk strategies experienced rapid drawdowns and failed to recover within the tournament timeframe.

Technical notes

The event served to compare behavioural effects of different algorithmic approaches rather than to validate deployable trading systems for live markets.

Results highlight the importance of risk management settings and demonstrate that isolated successful trades did not translate into net profitability for any entrant.