What will be true of OpenAI's Orion (GPT-4.5) model?
82
1.8kṀ9910
resolved Feb 28
Resolved
YES
The preparedness scorecard for the model will not be above Medium risk for any category
Resolved
YES
It will be released before Claude 3.5 Opus or Claude 4
Resolved
YES
It will cost more than 4o on the API ($10/1M output tokens)
Resolved
YES
It will cost more than 5x more than 4o ($50/1 million output tokens)
Resolved
NO
It will score better on SWE-Bench Verified than Claude 3.5 Sonnet (October version)
Resolved
NO
It will be called GPT-5
Resolved
NO
It will be able to output audio without calling another model
Resolved
NO
It will be able to output images without calling another model
Resolved
NO
It will be able to take video as input
Resolved
NO
It will have a context window of >= 1 million tokens
Resolved
NO
It will score better on GPQA than o1-preview (73% pass@1)
Resolved
NO
Once it is available to the public, a Manifold poll asking if it is better or worse than expected will find that it is better than expected
Resolved
NO
It will be released via ChatGPT before the full o1 model is released via ChatGPT (not o1-preview)
Resolved
NO
It will have a context window of >= 500K tokens

  • Update 2024-21-12 (PST): - If Orion is not planned for release, most options will be resolved as N/A (AI summary of creator comment)

  • Update 2025-02-13 (PST) (AI summary of creator comment): Resolution Update:

    • Release Confirmation: Orion is now confirmed to be released.

    • Resolution Timing: The market will resolve next month based on this confirmed release.

  • Update 2025-02-16 (PST) (AI summary of creator comment): Anthropic Model Naming Exception

    • If Anthropic releases its reasoning model before Orion, but it is not named Claude 3.5 Opus or Claude 4, the market will resolve YES.

  • Update 2025-02-27 (PST) (AI summary of creator comment): GPQA Option Update:

    • The resolution for the GPQA option is now set to NO due to a correction by the creator.

Get
Ṁ1,000
to start trading!

🏅 Top traders

#NameTotal profit
1Ṁ410
2Ṁ154
3Ṁ145
4Ṁ144
5Ṁ138
Sort by:

Why N/A on GPQA?

@gallerdude @mods Please reresolve the GPQA option to NO, I made a mistake

@SaviorofPlant it's now NO

@Ziddletwix thanks!

bought Ṁ168 NO

@SaviorofPlant How would this be resolved if Anthropic had released its reasoning model before Orion, but it would be not named Claude 4?

sold Ṁ20 YES

@JanPydych Would resolve YES. Don't really understand why it's trading so low, seems plausible the new Anthropic model will not be named either of those things?

@SaviorofPlant To be honest, I think that one of the options could be the release of a new checkpoint of Claude 3.5 Sonnet, but with the addition of the "reasoning_effort" parameter (or however Anthropic will name it).

@SaviorofPlant As I understand, according to the latest from Sam Altman, GPT-5 is planned to be a combination of 4.5 and o3, or something like that? Will probably N/A this option in that case.

bought Ṁ20 YES

What happens if it is not released by Jan 1?

bought Ṁ40 NO

@JoshYou I will extend the close date of this market until the release is announced.

If the release is delayed and it's unclear whether a released model is Orion, I'll wait for high quality reporting on whether or not a new model is Orion or not. If this never comes, every answer N/As (besides the "it will be released before X" options).

@SaviorofPlant Based on The Information articles, Orion is apparently not planned for release. Not sure how long I'll wait before N/Aing most of these

@SaviorofPlant Looks like it's being released after all, this market should resolve next month. Reopening for a week

© Manifold Markets, Inc.Terms + Mana-only TermsPrivacyRules