Meta’s Maverick AI Model Falls Short on Benchmark

• Earlier this week, Meta faced scrutiny for using an experimental Llama 4 Maverick model to score high on LM Arena.
• LM Arena maintainers changed policies and scored the vanilla Maverick, which ranked lower than rivals.
• The unmodified Maverick fell behind models like OpenAI’s GPT-4o and Google’s Gemini 1.5 Pro.
• Meta explained that their experimental Maverick was optimized for conversationality, skewing benchmark results.
• A Meta spokesperson expressed excitement for developers to customize Llama 4 and provide feedback.

Source: TechCrunch