Which of the following companies will release AI models in March?

12kṀ68k

resolved Mar 31

Resolved

YES

Alibaba (Qwen)

Resolved

YES

DeepSeek

Resolved

YES

OpenAI

Resolved

YES

Google

Resolved

YES

Cohere

Resolved

YES

Mistral

Resolved

Anthropic

Resolved

groq

Resolved

Grok

Market to determine which companies release AI models in March 2025. Note -- no distinction between reasoning, non reasoning and vision models. Criteria remains the same-- should be publicly accessible.

Update 2025-03-07 (PST) (AI summary of creator comment): Release Timing Clarification
- The market will count any model that is released in March regardless of the specific start date within the month.
- There is no distinction based on the day or timing of the release in March.

Update 2025-03-07 (PST): Research previews: Releases identified as research previews do not count. Model updates with date changes: If a release is only updated by a new date in the API, it is to be treated as a new model release. Discriminative models: These types of models will not be counted. (AI summary of creator comment)

Technology

Technical AI Timelines

OpenAI

AI Impacts

Get

1,000

to start trading!

🏅 Top traders

#	Name	Total profit
1		Ṁ2,704
2		Ṁ1,335
3		Ṁ1,321
4		Ṁ995
5		Ṁ764

People are also trading

Which of these model providers will sign the EU AI Act Codes of Practice?

AWS or Microsoft release own frontier AI model?

75% chance

Will OpenAI form a partnership with the following studios/companies for using models for AI generated videos? (By 2026)

What will be the top-3 AI tools in 2025?

Will AI models from company other than OpenAI comes first for a duration of 2 months by end of date?

In Jan 2027, it will be standard practise for non-AI-building tech companies to finetune and train their own models

Sort by:

people -- feel free to add more options to this!!

bought Ṁ550 YES

since dates count as new models https://huggingface.co/deepseek-ai/DeepSeek-V3-0324

bought Ṁ200 YES

does grok 3 image editing count?

@MingCat I would argue yes?

@MingCat fundamentally a new model which has different/better capabilities than the previous one

bought Ṁ250 YES

@typeofemale no API, and they don't specifically call this a new model - that's fine though right?

@MingCat I think so, I will resolve this an hour later if no one responds no (I am also not betting on this for fairness)

sold Ṁ272 YES

@typeofemale looks like this functionality existed for a while? i think this may be an update of the wrapper rather than the capabilities of the model - grok 3 image may be natively multimodal https://www.reddit.com/r/grok/comments/1j2c0kr/grok3_image_editing_is_underrated_its_insanely/

@MingCat interesting, then I won't resolve yes

@typeofemale @Bayesian thoughts?

none, havent followed this at all mb

https://openai.com/index/introducing-our-next-generation-audio-models/ Two new

speech recognition models and one new TTS model.

bought Ṁ50 YES

https://manifold.markets/typeofemale/1000-mana-for-prompt-engineering-th plugging a related bounty here in case someone wants to try

2000 mana bounty for prompt engineering this :)

Okay, I have a new test for all your "frontier models": can they answer the following prompt correctly "can you count how many pieces are there in this set cumulatively? extra points if you can guess what this is, and what parts are missing?" [image]Ofc, this prompt fails to give me a satisfactory answer. So -- can you give me a prompt which can answer this question? Feel free to break out your favorite reasoning models!! a) prompts which depend on searching the internet will receive a deduction (i'm not disqualifying them) b) the entirety of the prompt, along with any additional setting must be provided c) the output must be correct and be independently validated by either me (or someone else) d) I am intentionally not providing any information in this question so that people can't just feed the link and get an unfair advantage :). Any spoilers shall be deleted!!!!!

bought Ṁ150 NO

@rohanvisme what do you know? (ie why did u buy it up to 75%)

@Bayesian no insider information on my part, and most of my moves made on this site are done with very little actual thought.

The truth is ... I sense the vibrations in the fabric of the model-verse. The GPUs speak to me.

And like most of my GFs, they lie to me.

bought Ṁ150 NO

@rohanvisme Ty that makes sense

cohere apparently released a aya vision last week

groq doesn't train models to my knowledge, or do you mean adding other companies' models to their api?

@hecko I agree, I didn't add the option though

@typeofemale Do generative and discriminative models both count? How about updates to existing models only distinguished by a date in the API? How about research previews, what's the cutoff for publicly accessible?