Minor changes to the name, such as using "o4-large-1125" instead of "o4", are acceptable when determining the announcement. Moreover, skipping some numbers (o5, o9) is similarly acceptable.
To count as announced, an official announcement, with information about the model's capabilities, must be made by OpenAI or an official partner of OpenAI. if it's banned in the US but releases in China this market would still resolve YES.
Note that the model being publicly available at a future date is not required. This market, had it been about o3, would have resolved YES "before January 2025".
Integrated GPT-5 scenario: If o4 is simply incorporated into GPT-5 (i.e. if o4 is not independently announced), the market resolves NO.
Independent or benchmark announcement: If o4 is trained independently or on top of GPT-5 and there is an official announcement (e.g. benchmark results) even if it is not released as a standalone model, the market resolves YES.
See also:
/Bayesian/when-will-openai-announce-o4 (this market)
/Bayesian/when-will-openai-release-gpt45-ULpyUuqd2L
/Bayesian/when-will-openai-release-an-open-so
/Bayesian/when-will-deepseek-release-r2
1,000@TimothyJohnson5c16 fair, this market would resolve NO in that case. however if o4 is trained independently of gpt-5 or on top of gpt-5, in a way where they announce o4 benchmark results but say "btw it won't be relaseed independently, you can use it through gpt5", you agree that would count as openai announcing o4 right?