What will be true of OpenAI's Orion model?
➕
Plus
38
Ṁ3070
Nov 30
84%
The preparedness scorecard for the model will not be above Medium risk for any category
79%
It will score better on SWE-Bench Verified than Claude 3.5 Sonnet (October version)
75%
It will be able to output audio without calling another model
70%
It will score better on GPQA than o1-preview (73% pass@1)
59%
It will be able to take video as input
47%
It will be released before Claude 3.5 Opus or Claude 4
45%
It will be called GPT-5
44%
It will have a context window of >= 1 million tokens
36%
It will be able to output images without calling another model
32%
It will be released via ChatGPT before the full o1 model is released via ChatGPT (not o1-preview)
19%
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

️ TechnologyAIOpenAITechnical AI Timelines️ AI Alignment
Get Ṁ1,000 play money
3 Comments
Sort by:
bought Ṁ20 It will be able to t... YES

What happens if it is not released by Jan 1?

bought Ṁ40 It will score better... 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.

bought Ṁ150 It will be able to t... NO

https://www.theverge.com/2024/10/24/24278999/openai-plans-orion-ai-model-release-december

Related questions