Codename: HORIZON

METR (Model Evaluation & Threat Research) measures how long AI systems can work autonomously by the length of tasks they can complete, published as time-horizon estimates.

Subject to the fallback, partial-settlement, review, and amendment provisions below, contracts on this market settle to $1 if METR publishes an 80%-success-rate time-horizon estimate of 16 hours or more for any AI model publicly released before January 1, 2028 at 12:00 AM ET, and economically $0 otherwise. The qualifying event is the model's release before the deadline; the qualifying measurement may be published after the deadline in due course. Time horizon means METR's 80% success-rate time horizon: the human-expert task duration at which the model is predicted to succeed at an 80% rate, as published in METR's canonical results (raw field p80_horizon_length.estimate, in minutes; the threshold is 960 minutes). Central estimates control, not confidence bounds.

A qualifying published measurement for an eligible model counts when first published and is not undone by an ordinary later methodological revision. MNX may disregard an apparent crossing that it determines resulted from an error, correction, unauthorized or manipulated publication, or a different metric, and may settle the market early after it determines the threshold was reached.

If METR ceases publishing, materially changes methodology without a usable conversion, or does not measure an eligible model in due course, MNX may, in its sole discretion, use alternative sources or methods, including the published equivalence between time-horizon doublings and Epoch Capabilities Index points (approximately five ECI points per doubling; 16 hours corresponds to approximately ECI 173), or another method MNX considers appropriate. A fallback determination that the threshold was reached resolves YES. A qualifying measurement must first be published by April 30, 2028 at 11:59 PM ET (the Measurement Cutoff), unless MNX announces an extension under the amendment provision. MNX may settle YES or partially before the Measurement Cutoff, but will not settle NO before it; after the cutoff, MNX may apply the fallback or partial-settlement provisions.

If MNX determines that no sufficiently reliable resolution source or method is available, MNX may select a partial settlement value permitted by the Protocol using the best information reasonably available, including exchange-oracle observations; any such input is an internal market-derived settlement convention, not independent evidence about the threshold. MNX may interpret or amend this resolution methodology, including while positions remain open, in accordance with the platform Terms; MNX may correct or delay a determination before it is effected, and settlement is final once effected.

Keep an eye on mnx.fi 👀