GPT-5 capabilities at launch - make your predictions
92
822
4.4k
2031
48%
to be N/A'd : Seed answer to create a market
69%
Video modality(Input or Output)
96%
Audio Modality(Input or Output)
96%
Image Modality(Output)
93%
Context size >= 200k
73%
Context size >= 500k
44%
Context size >= 1000k
31%
Context size >= 2000k
33%
Online learning(agent finetuning, PEFTs, etc.)
53%
Exact query of custom knowledge datasets(e.g. Google RETRO, where users can provide their own datasets)
7%
Designed to inference cheaper than GPT-4(price will be lower)
71%
Able to solve easy Sudoku puzzles with prompting
33%
New Neural Network Architecture
19%
Capable of online learning
35%
Able to precisely count the amount of triangles there is in a image of non-overlapping radomly placed shapes(for images with less than 50 shapes)
41%
Would qualify for AIME given screenshots of the AMC 10 questions, from the AMC 10 released soonest after GPT-5 release

Since others can add in their predictions, I will use a combination of their opinion about resolution and my subjective judgement to resolve the predictions. I will not trade (other than to make token predictions with Ṁ1 ).

For resolution of predictions: Resolves YES to those that are more or less correct (in spirit) or to 50% (if technically true but not in spirit) or to NO (if not true). N/A if irrelevant.

If GPT-5 is not launched this decade, market's state will be resolved to N/A. You can trade on whether or not that happens here:

Get Ṁ600 play money
Sort by:
Able to solve easy Sudoku puzzles with prompting

uhh with prompting? Like, Mira-challenge style?

New Neural Network Architecture

@Shump or @firstuserhere how do you plan to resolve this if they don't speak much to the architecture?

@RobertCousineau I'm going to defer to the prediction's author (here @Shump ) by default for this market unless controversy ensues/author inactive/ doesnt clarify

etc

@RobertCousineau @firstuserhere I'd change the description to New Architecture Announced and resolve NO if they don't say anything. Can N/A on that but that just seems less interesting.

I wrote this months ago and honestly I now regret something so vague, but I'd define new as something substantially different than the current norm. So for example replacing tranformers is an obvious YES, while using Flashattention, which is just a different implementation is NO. I am still here so I can rule on it if needed.

bought Ṁ30 New Neural Network A... NO

@Shump Wdym less interesting? Isn't N/A the standard of doing things in such cases? If there was a different market on whether they'll disclose details about their new models, I'd bet no.

reposted

reposting because the launch may be near

bought Ṁ10 of Context size >= 1000k YES

image modality says output. is that also what audio/video modality means?

bought Ṁ1 of Able to solve easy S... YES

@jack No. Audio/video modalities can be input or output. But GPT-4 already has image input, so it would be nothing new if GPT-5 also does image input.

Video is also a different thing from "ability to take multiple images as input".

@jack I've edited this into the answers since it seems they added the capability.

Somewhat related

Able to solve easy Sudoku puzzles with prompting

lul it would be legendary if they address this