Resolution criteria

This market will resolve to YES if at least two (2) of the top five (5) AI models ranked on the Epoch Capabilities Index (ECI) general leaderboard are open-weight or open-source as of December 31, 2027. Otherwise, this market will resolve to NO.

• Primary Source: The official Epoch Capabilities Index leaderboard at epoch.ai/benchmarks/eci.

• Top 5 Ranking: Ranking is determined by the ECI general capabilities score. If multiple sizes or post-training variants of the same model series are listed (e.g., a base version and an instruct version, or different parameter sizes of the same model line), only the single highest-scoring version will be counted toward the top 5.

• Definition of Open Source / Open-Weight: A model qualifies if Epoch AI classifies its weight accessibility as "Open weights" or "Open source," or if its weights are publicly downloadable for local inference (e.g., Meta's Llama series, Alibaba's Qwen, DeepSeek, or Mistral models), even if released under custom commercial-use licenses.

• Fallback: If Epoch AI stops updating the ECI, or if the website is permanently offline in late 2027 with no official snapshot or replacement benchmark published by Epoch AI, this market will resolve N/A.

Background

The Epoch Capabilities Index (ECI) is a composite benchmark developed by Epoch AI that aggregates performance across dozens of distinct evaluations (such as FrontierMath, ARC-AGI, and others) into a single unified capability scale. Because individual benchmarks frequently suffer from saturation as models improve, the ECI utilizes a statistical framework to trace holistic AI progress over multiple years.

While closed-source frontier models from OpenAI, Anthropic, and Google DeepMind have historically held the top positions on the ECI, open-weight models from Meta, Mistral, and Chinese developers like Alibaba (Qwen), DeepSeek, and Moonshot AI (Kimi) have steadily closed the capabilities gap. This market tracks whether open-weight models can secure a multi-seat presence at the absolute frontier of AI performance by the end of 2027.

This description was generated by AI. Review and verify everything here yourself. You can edit, replace, or delete any part of this description, including the resolution criteria. You do not need to trust the AI output.