Claire Vo · 2 min read · Gen AI Tools
Meta’s new ‘Muse Wonder’ model: good for AI agents, bad for design
Read the original on lennysnewsletter.com →
Claire Vo benchmarks Meta's new Muse Wonder and Wonder Full frontier models across nearly 20 tasks including prototyping, agentic workflows, coding, writing, and creative work using her 'How I AI' bench. The evaluation reveals strong performance in certain agentic tasks but notable weaknesses in design output. The models are assessed against competitors like Sonnet, Opus 5.5, Astra, and Fable across various use cases.
Key Takeaways
- Muse Wonder ranked first in three agentic task categories, making it potentially valuable for agent workflows despite some limitations
- Design output from Muse Wonder is poor quality, making it unsuitable for prototyping and visual design tasks
- Muse Wonder is fast on one-shot tasks but slower on complex agent tasks compared to alternatives
- Performance on SVG illustration and short-form video editing lags behind competitors like Opus 5.5, Astra, and Fable
- Meta's 'personal and professional agent' framing mostly holds up, though the model's 'active focus mode' for long-running tasks has mixed real-world results
Excerpt
I put Meta’s new Muse Wonder (and Wonder Full) frontier models through the How I AI bench, which covers nearly 20 tasks across prototyping, agentic workflows, coding, writing, and creative work. I scored everything with an LLM-as-a-judge pass and my own voice notes, then pulled together the first-look results you see here. Listen or watch on YouTube , Spotify , or Apple Podcasts What you’ll learn: Why Meta’s “personal and professional agent” framing mostly holds up, and where it doesn’t What active focus mode actually does on long-running tasks, and whether it’s real The three agentic tasks where Muse ranked…