Which company has best AI model end of March? (Chatbot Arena Leaderboard)
197
1kṀ740k
resolved Apr 1
100%99.0%
Google
1.0%
OpenAI
0.0%
xAI
0.0%
Anthropic
0.0%
Meta
0.0%
Alibaba
0.0%
DeepSeek
0.0%Other

Resolution is based on the chatbot arena LLM leaderboard (https://lmarena.ai), specifically the company with the highest Arena Score in the Overall category without filters (without style control or show deprecated), at the end of March 31, 2025 11:59PM ET.

See also:

Get
Ṁ1,000
to start trading!

🏅 Top traders

#NameTotal profit
1Ṁ7,221
2Ṁ2,331
3Ṁ1,792
4Ṁ954
5Ṁ688
Sort by:

fucking brutal

@MalachiteEagle eh it's cope. There are two labs above OA on lmarena. Sam's days are numbered.

@jim haha guess we'll see. I think OpenAI has some internal results with o4 (or whatever they're calling it) that are quite far beyond anything Google has.

@jim oh and the image stuff is just with gpt-4o, will be interesting to see how fast they ship gpt-4.5 image capabilities

@MalachiteEagle is it confirmed that 4.5 can generate images natively? i thought 4o was unique in its multimodality?

@SaviorofPlant I haven't seen them explicitly confirm it, but I think the likelihood is very high, especially since it was originally intended to be their gpt-5 model

@MalachiteEagle i think 4o was trained more recently than 4.5, which has an october 2023 knowledge cutoff. seems likely to me that 4o was an experimental model testing this type of multimodality and previous models did not have it

@SaviorofPlant they have a fine-tuning/continual-learning technique allowing them to update the knowledge cutoff from a given base model

@SaviorofPlant they got multimodal working with gpt-4o, released it (without the multimodal output) and then started the work / training run for gpt-4.5. This produced lower-quality results than they hoped, so they went in circles for a few months until the episode with Q* happened

@MalachiteEagle oh hang on, that's not right. They released gpt-4o in first half of 2024. Q* shenanigans happened in late 2023

@SaviorofPlant but point being that I'm pretty sure gpt-4.5 training run was a scaled up version of what they got working for gpt-4o

boughtṀ550 YES

@Bayesian Nice snag. Now go beat Hillary in Masters. 😆

bought Ṁ15 NO

@Quroe i might make more from losing we'll see

nvm not really significant

bought Ṁ11,075 NO

@jim you’re buying this at 2% when it’s available at 0.1c on polymarket

in other words: not gonna happen

bought Ṁ100 NO

@Frankas impressive but doesn't seem likely to lead to a rank 1 placing

@jim their "max" model is supposedly coming soon

https://x.com/Alibaba_Qwen/status/1894130603513319842

opened a Ṁ3,000 YES at 50% order

Wouldn't they already be testing these models anonymously on lmsys arena if they were to take the lead there by the end of the month?

I take that to mean there's a new anonymous model on lmarena 🤔

Haven't seen anyone mention it on twitter yet though

@MalachiteEagle Not necessarily no, there’s 40 days left

Oh sorry I misread that as "before March".

Here's a question in case one of these appears tomorrow: https://manifold.markets/MalachiteEagle/will-there-be-a-new-and-smart-anony?play=true

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