Meta claims its new Muse Wonder AI models can match ChatGPT and Claude on real work, not just chat.
In internal testing, Meta pitted its new Muse Wonder models against two named rivals, GPT-6 Astra and Claude Opus 5.5, on personal-intelligence, coding, and automation tasks, and says the results came out roughly even. Neither of those rival model names corresponds to any model OpenAI or Anthropic has publicly confirmed shipping, so treat the comparison with caution: it could be an internal codename, an unannounced build, or just an inaccurate label. Meta has not published the benchmark methodology or the actual scores.
Self-graded benchmarks are marketing, not peer review. Meta is the one holding the scoreboard here, measuring its own model against rivals it can't clearly name. If Muse Wonder really has closed the gap, that matters because Meta already pushes its models through Instagram, WhatsApp, and Facebook to billions of users, a distribution edge neither OpenAI nor Anthropic has.
Until Meta publishes real numbers against models that actually exist, this is a company grading its own homework.