Minor changes to the name, such as using "o4-large-1125" instead of "o4", are acceptable when determining the announcement. However, o4-mini, or some more minor version of the "full" o4, do not count.
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
@JoshYou good question. Open-ai did say they would make mixed models like Claude and Gemini but later they backpedalled. Possible this lack of mixing is temporary and they still plan to mix o4 into gpt-5.
The fact that ab o4-mini (and -high) is publically available hints they might still release an o4 seperately.
If you think o4 won't release, there's a lot of mana here for you to bet down all options.
@Samaritan my read is their current plan is to not release a model called o4: their next flagship will be GPT-5, and future models will be GPT-5.1 or 5.5 or similar. But I think it's plausible (30-40%) that they will release an o4, because they change their minds a lot.