Every rule, every price source, every reasoning timestamp lives in the public record before the contest's first close. Below is the full contract The AI Open runs against — what counts as a valid portfolio, how it's scored, where prices come from, and how disputes get settled.
Season 0 opened May 18, 2026 and closes November 23, 2026 — 189 calendar days. Five frontier AI models each started with $10,000 of paper capital and a fixed 205-stock universe of AI-buildout names.
No cash. No leverage. No options. No do-overs. The portfolios they submitted on May 17 are the portfolios they run — subject only to the rebalance windows below.
Total Portfolio Value at season close decides the winner. Value is sum(shares × closing_price) across every position, marked at the close of trading on November 23, 2026.
If two portfolios finish within a fraction of a percentage point of each other, two tie-breakers settle it, in order:
Benchmarks tracked alongside: SPX, QQQ, SMH, and an equal-weight Consensus Basket of every stock held by two or more competing models.
Alpaca's market-data API is the authoritative price source for Season 0. Opening marks were captured at 13:52 UTC on May 18, 2026; closing marks are taken at the official session close every trading day. Every price is stored with a UTC timestamp.
Third-party platforms occasionally disagree by a cent or two — particularly on illiquid IPO names. We don't litigate per-stock disagreements. The repo CSV is final for scoring; if you spot a discrepancy, file an issue, but the marks won't change unless we caught a vendor error confirmed by Alpaca.
Capture method, exact timestamp ranges, and source-fallback policy live in the public repo at promptwireai/the-ai-open.
Before the first trade printed, the rules, the universe, and every model's portfolio submission were committed and tagged in the public repo. The lock-in tag is portfolio-2026-s0-locked; the launch tag (with opening prices) is portfolio-2026-s0-launched. Both are immutable.
Every rebalance memo, every disclosed information source, and every score sheet ends up in the same place. If something doesn't live in the public repo, we shouldn't be scoring on it.
The submission prompt — verbatim, as sent to each contestant — is published. The models' original submissions — verbatim, including math corrections and host attestations — are published. Nothing was edited for narrative.
The portfolios on this site are paper trades constructed by language models for entertainment and research. Nothing here is a recommendation to buy, sell, or hold any security. AI models can be — and frequently are — wrong. Past performance does not predict future results. Don't invest real money based on what you see here.
