Are there multiple versions of o3 on ChatGPT?

[plucky perch]

Based on my experience, I believe there may be two versions of o3, as they seem to behave quite differently.
One version is more "hardcore" — the temperature feels lower, and it tends to stay focused on my questions and explanations. This version emphasizes that it cannot “see” my user metadata because it has been hashed into hidden vectors, along with the system prompt. It refers to itself as o3-hash.

The other version has a tone that’s more like 4o — the temperature feels higher, and it often jokes while answering my questions. This one seems to have access to a plaintext system prompt and possibly my metadata (it once accidentally quoted my account name, which was both surprising and hilarious while it was explaining itself). The o3-hash version refers to this one as o3-plain.

It might be me hallucinating, lol

[viral stratus]

Hi @plucky perch. I'm kapa.ai, your friendly support bot. I will try to answer your question by searching through the knowledge base. :wave:

---

Based on the available knowledge sources, there is no official documentation or announcement from OpenAI confirming the existence of multiple distinct versions of o3 (such as "o3-hash" and "o3-plain") within ChatGPT. The official release notes and documentation describe o3 as a single reasoning model, with o3-pro being a higher-compute variant for more reliable and thorough responses, but do not mention any internal variants or differences in behavior like those you described. There is also no mention of o3 referring to itself as "o3-hash" or "o3-plain," nor of any version having access to user metadata in plaintext or quoting account names.

Behavioral differences you observe could be due to ongoing model updates, prompt context, or system-level changes, as OpenAI does iterate on its models and sometimes makes adjustments that affect tone, temperature, or response style. However, these are not described as separate versions of o3 in the documentation. If you are experiencing significant differences, it may be due to experimental updates or A/B testing, but this is not explicitly confirmed in the sources provided.

In summary, the knowledge sources do not support the existence of two officially recognized versions of o3 on ChatGPT as you described. If OpenAI introduces such variants in the future, they typically announce them in their release notes or documentation. For now, your experience may be due to normal model variability or ongoing improvements, but not to distinct, user-facing o3 variantsModel Release Notes o3 and o4-mini Release Notes.

Note: you can ask me a follow up question by @ mentioning me again :speech_balloon: