Twelve dimensions across seven providers. Re-graded as announcements land, with the market's reaction shown alongside.
Raw capability · updated 2026-06-01
Llama is strong and improving, a notch below the very top frontier models.
How smart is the assistant at the core reasoning, knowledge, and generation tasks people actually throw at it? Frontier benchmark standing, but weighted toward real assistant quality rather than leaderboard maxing.
Profiles the ~50-person TBD Lab at the core of Meta's superintelligence reorganization.
Each giant's AI strategy flows from its computing philosophy; Meta's 'personal superintelligence' is augmentation.
The Meta AI app, built on Llama 4, leans on Facebook/Instagram profile data for personalized assistance.
Meta submitted an experimental chat variant to LMArena that differed from the public weights, a credibility hit.
Zuckerberg's four AI bets, and the argument that Meta spends enormously yet lacks a coherent direction.