Criteria:

• the model must be OpenAI's most capable public model at launch

• it must be able to process audio natively, in a similar way to how images are processed today, not through tools

• It must demonstrate rudimentary understanding of non-speech audio

gpt-4o would have met the criteria at the time, but it has since been superseded. This market resolves when a new model is released fitting these criteria, in or after 2026.