Is GPT-4.5 the base model for o3?

1kṀ4536

2031

10%

chance

ALL

This market resolves to YES if OpenAI makes an official statement or announcement that confirms GPT-4.5 is the base model for o3. It resolves to NO if OpenAI makes an official statement or announcement that confirms a different model (not GPT-4.5) is the base model for o3.

Resolution Criteria

Resolves YES if:

OpenAI explicitly states in an official communication (blog post, press release, technical paper, official social media account, or statement from leadership) that GPT-4.5 is the base model for o3
The statement clearly identifies GPT-4.5 as the foundational architecture or model that o3 is built upon
If o3 is described as "built on," "derived from," or "a version of" GPT-4.5

Resolves NO if:

OpenAI explicitly states in an official communication that a model other than GPT-4.5 (e.g., GPT-5, GPT-4o, a completely new architecture) is the base model for o3
OpenAI explicitly denies that GPT-4.5 is the base model for o3

Evidence Standards

Only official OpenAI sources will be considered as valid evidence
Unofficial leaks, rumors, or third-party analysis will not be considered for resolution
Technical analyses by outside parties attempting to determine the base model will not be considered for resolution

Resolution Timeline

This market will resolve when OpenAI makes a clear statement about o3's base model.

OpenAI

LLMs

Get

1,000

to start trading!

People are also trading

Will GPT-5 be GPT-5o?

90% chance

What is the main reason behind GPT-4o speed improvement relative to GPT-4 base model?

Is GPT-4 best? (Thru 2025)

63% chance

How many parameters does GPT4o have?

Will we have an open-source model that is equivalent GPT-4 by end of 2025?

82% chance

If we find out in 2024, was o1's Transformer base trained on 10+x as much compute as GPT-4's?

19% chance

GPT-5 exists

47% chance

Will training on o1/o3 traces improve GPT-5 performance on a broad set of tasks?

63% chance

GPT-4 #5: Will GPT-4 be a dense model?

Sort by:

I think this is unlikely:

o3 is only about three months behind o1, based on when they announced benchmarks results (September and December). o1 release and Deep Research were also three months apart. This means they moved pretty quickly to train o3 after training o1. Orion might not have been ready in time, and in any case switching base models probably requires time to adjust the reasoning training recipe.
GPT-4.5 is big and expensive, which would make inference and training for o3 more expensive

Also if o3 was based on 4.5 they probably would have bragged about that fact instead of being relatively modest about 4.5's capabilities.

bought Ṁ100 YES

Seems likely based on this only mentioning o3 mini, with the reference to "future reasoning models" being based on 4.5

You may want to edit the title to read "Will OpenAI say GPT-4.5 is the base model" to better align with the resolution criteria. It could be proven beyond reasonable doubt that it is, yet OpenAI won't confirm for business reasons.

@WilliamGunn I mean if OpenAI doesn't confirm it won't resolve No

@NoaNabeshima Right, and potential traders need to know you're relying on OpenAI's word. This isn't clear from the question wording.

@WilliamGunn Idk, that also feels misleading (and more so than the current title to me) because the market won't resolve NO if OpenAI doesn't say that GPT-4.5 is the base model. I don't feel too worried about this, I think the resolution criteria in the description is pretty clear and not very different from the title. I might eat my words later, of course, we'll see.

@NoaNabeshima A company's statements and the fact of the matter are often different, and you are resolving on statements but the question asks about the fact of the matter. You're putting traders in a situation where GPT4.5 could be shown by a credible third-party to be the base model but you resolve no because OpenAI didn't say anything. You can avoid this by asking instead "Will OpenAI confirm GPT4.5 is the base model for o3?".

You can deliberately mislead people or change the wording of either the question or the resolution criteria, but you can't pretend there's no issue with it as is.

@WilliamGunn Notably I would not resolve No if credible third parties show that GPT4.5 is the base model and OpenAI doesn't say anything. I am not putting traders in that situation.