#Llama 4
2 Stories
Meta Under Fire for Manipulating Llama 4 Benchmark, But It Isn’t the First Time
Arjun Sha
View quick summary
Meta is facing accusations of hacking the Llama 4 benchmark score, especially on Chatbot Arena, by blending benchmark test sets during the post-training phase. We break down the controversy and highlight how this isn't the first time Meta has been caught gaming the benchmark.
Meta has unveiled a new series of Llama 4 open-weight models, and they look impressive. The Llama 4 Maverick model outperforms GPT-4o and Grok 3 in the LMArena benchmark. All Llama 4 models are developed on the MoE architecture and they are natively multimodal.