- Meta has released Muse Wonder and Wonder Full, new AI models that read and respond to text. According to the source, Muse Code subscribers can use them starting that day, and Wonder is also accessible to developers through the Meta Model API.
- Meta claims that both models are comparable to OpenAI's GPT-6 Astra and Anthropic's Claude Opus 5.5 and Claude Fable 5.1. The basis is several industry benchmark results from Meta's internal tests, and the source does not include 3-party verification results.
- Meta said it plans to power its Muse agents with Wonder and to build specialized models in the future. It said the change would first apply to subscribers, but did not disclose specific details or a timeline.
- According to the source, a performance comparison with Google's Gemini 4 Argon is not yet clear.
The first thing that catches my eye is that Gemini 4 is left out of the Argon comparison. The advantage Meta claims comes from its own internal tests, so from the original text alone it is hard to tell which tasks and settings were used. For those who use Muse Code, I wonder whether you have put the work you usually hand to it side by side with existing models and Wonder?