Meta expects that, starting in 2025, future Llama models will become the most advanced AI models in the industry.

The test is whether a Llama model released in 2025 was the most advanced AI model in the industry overall, judged against all frontier models (closed models included) on widely cited leaderboards and benchmarks in 2025, such as LMArena overall rankings, the Artificial Analysis intelligence index, and standard reasoning, coding and maths benchmarks. True: a 2025 Llama model ranked first overall, or tied for first, among all frontier models on most of these measures, for more than a brief period. Mostly true: a 2025 Llama model was at the top overall on some major leaderboards and within a few positions of the top on the rest, so it was broadly a co-leader. Mixed: it led only in a few named categories or on some benchmarks, while trailing the leading closed models overall. Mostly false: Llama led only among open-weight models, which drops the claim's 'in the industry' qualifier, or it reached the top only after 2025. False: no 2025 Llama model ranked at or near the overall top, for example by clearly trailing the leading models from OpenAI, Google or Anthropic on major leaderboards.

Set before the research began.

1. E1 Meta's vanilla Maverick AI model ranks below rivals on a popular chat benchmark | TechCrunch TechCrunch · 11 April 2025

   “The unmodified Maverick, “Llama-4-Maverick-17B-128E-Instruct,” was ranked below models including OpenAI’s GPT-4o, Anthropic’s Claude 3.5 Sonnet, and Google’s Gemini 1.5 Pro as of Friday.”

2. E2 Meta's benchmarks for its new AI models are a bit misleading | TechCrunch TechCrunch · 6 April 2025

   “One of the new flagship AI models Meta released on Saturday, Maverick, ranks second on LM Arena, a test that has human raters compare the outputs of models and choose which they prefer. But it seems the version of Maverick that Meta deployed to LM Arena differs from the version that’s widely available to developers.”

3. E3 raw AI Wiki (aiwiki.ai)

   “When the unmodified public release of Llama 4 Maverick was added to the leaderboard, reports indicated it ranked far lower, around the thirty-second position, well below older models from competing providers.”

The evidence shows that the unmodified public Llama 4 Maverick ranked below leading OpenAI, Anthropic and Google models on LM Arena in April 2025 (E1, E3), and that a different Maverick version deployed to LM Arena ranked second (E2). It does not cover other 2025 Llama models or the other leaderboards and benchmarks the criteria name, such as the Artificial Analysis index and reasoning, coding and maths benchmarks. So it does not show that no 2025 Llama model ranked at or near the overall top.