Publicly available: Must be available to normal users (paid ok) via API, chat, or similar at some point before 2028.
General purpose: This is based on advertising and positioning of the model. GPT-5.3-Codex would not count because it's positioned as a Codex-specific model, as evidenced by its exclusive availability within Codex. GPT-5.2 would count. In general if the model is released within a chatbot, that's sufficient to make it general purpose.
Major AI Labs: For this market, this list includes OpenAI, Anthropic, Google, xAI, and DeepSeek. This list won't change.
Advertised with robotics capabilities: The model's official announcement or official blog posts on the same day as the model's announcement must specifically mention the model's robotics capabilities. This could be by mentioning that the model was trained on robotics trajectories, including a robotics benchmark result, showing videos of the model controlling a robot, etc.
1,000@ChurlishGambit For the robotics capabilities specifically that's correct. The general purpose model which is claimed to have them must at least exist and be publicly available as described.
I think it would be great also to have a market which tests the robotics capabilities of such a model practically, but I don't know of any authoritative benchmarks for such capabilities.