Resolves YES if, on or before 2026-12-31, any frontier AI lab (OpenAI, Anthropic, Google DeepMind, Meta AI, xAI, etc.) publicly discloses or reliably reports a single training run for a frontier-class model where the total compute cost exceeds $1 billion USD.

Acceptable evidence for resolution:

• Direct disclosure by the AI lab in a public technical report, model card, blog post, or earnings call

• Reliable third-party calculation by Epoch AI based on disclosed FLOP count + market GPU-hour pricing

• Peer-reviewed academic paper or arXiv preprint with credible cost estimate

The training run must be for a single model (not a research program total). Cost includes GPU-hours, not just OpEx; published estimates of "$X to train" are accepted at face value if from credible sources.

Source of truth: Epoch AI's tracking data, AI lab public communications, peer-reviewed academic publications.

Resolution date: 2026-12-31.

About this market. This market is part of SCB/SCO Reference Run #001 — AI Compute (30-day longitudinal demonstration) under Leadership Under Uncertainty. It is a research-demonstration corpus, not a commercial product.

Open-Sources-Only Commitment. This market resolves only against publicly accessible sources (SEC filings, government data, public benchmark publications, public corporate communications). Subscription-gated analyst content is not used in resolution.

Creator-is-not-trader. The creator of this market does not place trades on it. Probabilities reflect community trading.

Pre-registration. Question wording, resolution criteria, and close date were locked before any market data was observed and cryptographically anchored on day 0 of the run via the SCB/SCO daily Merkle seal. See protocol.