Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Previously, Meta this week landed in hot water Llama 4 Maverick model is an experimental, for use of the rolling version, LM Arena, LM Arena. Event He asked LM Arena’s protectors to apologizeChange their policies and hit the vanilla maverick.
It turns out, not very competitive.
Unrequited Maverick, “Llama-4-Maverick-17B-128E-128E instruction” The below was ranked in models Openai’s GPT-4O, Anthropic’s Claude 3.5 Sonnet and Friday, including Google’s Gemini 1.5 Pro. Most of these models are months.
The release version of the Llama 4 was added to Lmarena, but you probably did not see it, because you have to scroll to the 32nd place pic.twitter.com/a0bxkdx4lx
– P: ɡSN (@pigeon__s) April 11, 2025
Why poor performance? METAN’s experimental Maverick, Llama-4-Maverick-03-26-experience, “optimal for negotiations”, the company told a Published the diagram last Saturday. This optimization, human rankings played the LM Arena, which compares the performances of models and chooses the choice of choices.
As we write beforeFor various reasons, the LM Arena has never been the most reliable size of the AI model performance. Again, it makes it difficult to accurately predict how good the developers will be in different contexts – to make a model for a criterion.
In a statement, a meta spokesman told Techcrunch to the “all kinds of special variant types” with meta experiences.
“‘LLA-4-Maverick-03-26-experienced’, a fame that LMarena has also implemented the experience,” said Press Secretary. “Now we have left our open source and see how developers customize the Llama 4 for their use.
