What will be true of OpenAI’s open-weight model?

1.2kṀ3198

Dec 31

95%

Is a reasoning model

Is at least an Image-Language Model (ala JanusPro or 4o)

73%

Is at least a Vision-Language Model (ala PaliGemma)

73%

Uses Beyer Teacher/Student distillation ala Gemma 3

30%

my friend @soaffine (who knows ab ai) thinks it unveils an impressive new architecture

86%

An instruct model is made accessible

15%

A base model is made accessible

19%

The model is accessible to all users on the main chatgpt site

17%

Gets 1400+ Elo on lmarena.ai

14%

At least 300B parameters

24%

At least 70B parameters

78%

At least 30B parameters

72%

better than o3-mini on FrontierMath (with tools)

33%

Uses Mixture of Experts

58%

MIT or Apache license

Context:

Feel free to add your own answers. If an answer is unclear you are welcome to ask for clarification from both me and the person who submitted the answer.

OpenAI

Technical AI Timelines

AI Impacts

LLMs

Get

1,000

to start trading!

People are also trading

What will be true of OpenAI's Sora* model, at the end of 2025? [*see description]

Will OpenAI release a model which generates images using reasoning / inference-time scaling before 2026?

50% chance

Will OpenAI offer a model that updates its weights while running during 2025?

14% chance

When will OpenAI release an open source model?

OpenAI to release model weights by EOY?

87% chance

Next model open sourced by OpenAI?

Will the state-of-the-art AI model use latent space to reason by 2026?

19% chance

Will OpenAI Release a Model Capable of Reliably performing Gradeschool Math from Reasoning by Jan 1, 2025?

79% chance

What will be true of OpenAI's best LLM by EOY 2025?

Will OpenAI allow near full access to the weights of their best-trained model to an external auditor by the end of 2030?

Sort by:

"reasoning model" will be judged with reference to OpenAI documentation.

bought Ṁ10 YES

What’s the difference between image/language and vision/language?

bought Ṁ10 NO

@KJW_01294 I think image/language generates images whereas vision/language only takes them as input

I've chosen to bet on the market that references me, but assert that I will be truthful in my evaluation of it. To help others, I would have said no to Gemma 2/3, yes to PaliGemma, yes to DeepSeek-MoE, but no to DeepSeek V3 and DeepSeek R1