Grok 4.3

if still its a 500b model as elon claims, its not bad

It's a small model, pretty good actually.

basically MiMo V2.5 Pro level but much faster tps. - notably MiMo is open source, grok isnt.

4.3 is still a 500b checkpoint... Might be the best 500b model out I guess or ? Correct me if I am wrong. What size is the 6 models above Grok 4.3 ? [https://artificialanalysis.ai/models#intelligence-tabs](https://artificialanalysis.ai/models#intelligence-tabs) edit : It is as smart as Sonnet 4.6 and 5x cheaper and faster ...hmm

The free deep goon tier had a lot of people "excited" 🤭 Colossus 2 is scheduled for May 13 and will make xAI the AI compute leader until OpenAI goes Super Saiyan next year [according to this](https://epoch.ai/data/data-centers?view=graph&tab=power&mode=top-1&colorCategorization=primaryUser&colorPinned=OpenAI)

I'm a music producer and I always like to ask new released models to make some type of lyrics, cause, for some reason, AI can make magical things but can't write decent/non generic lyrics 😂 If you happen to try some lyrics, any type you want, I would love to see them. Also something I like about grok is the lack of censorship it has.

This is nice actually if it holds: > At 209 tokens per second, Grok 4.3 is notably fast. ~ Artificial analysis

I wonder how it’s gonna perform on alpha arena cause 4.2 is whooping the rest of the models currently

I generally prefer other companies, but the one area grok has excelled at for me is searching obscure information, specifically gaming related info. If I'm trying to figure out something specific about some shareware game from the 90s, I go to grok. Idk if other people have had similar results with grok.

havent pushed it too far, it seems dynamic vocabulary miiiiight be increased, but it’s difficult to really test without weeks of usage. However, one thing I noticed is it is able to discuss more controversial topics without hard-blocks now as it is able to better reason your intention (i.e. user isnt being harmful, they are exploring xyz topic in a safe manner) Not wildly different but definitely a slight increase. I use the chat for psych lit research and yes. I tested it in deciphering some heavily layered poetry as well but it didnt score much better than 4.20. So I assume 4.3 is just 4.20 with more training