Meta’s new ‘Muse Wonder’ model: good for AI agents, bad for design

Meta's new frontier models Muse Wonder and Muse Wonder Full ranked first on three agentic tasks in the How I AI bench, a roughly 20-task evaluation spanning prototyping, agentic workflows, coding, writing, and creative work, but produced poor design output, according to host Claire Vo's first-look results. Vo scored the models with an LLM-as-a-judge pass plus her own voice notes, and found Muse fast on one-shot work but slow on agent tasks, trailing Sonnet on voice assistant tasks and Opus 5.5, Astra, and Fable on SVG illustration and short-form video editing.

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 https://youtu.be/dChiPsd0uF4 , Spotify https://open.spotify.com/episode/26i50r5Xbbm5Tg1TeLXnqQ , or Apple Podcasts https://podcasts.apple.com/us/podcast/metas-new-muse-wonder-model-good-for-ai-agents-bad-for-design/id1809663079?i=1000796296040 What you’ll learn: 1. Why Meta’s “personal and professional agent” framing mostly holds up, and where it doesn’t 2. What active focus mode actually does on long-running tasks, and whether it’s real 3. The three agentic tasks where Muse ranked first, and why those categories matter 4. What Muse’s design output actually looks like it’s not good 5. How it compares to Sonnet on voice assistant tasks 6. Where Muse lands on SVG and short-form video versus Opus 5.5, Astra, and Fable 7. Whether I’m actually putting Muse into any of my agent workflows Brought to you by: DX https://getdx.com/howiai —Engineering intelligence for the AI era Jira AI SDLC https://ad.doubleclick.net/ddm/trackclk/N320202.6053656HOWIAI/B36397005.454541156;dc trk aid=648366078;dc trk cid=262806969;dc lat=;dc rdid=;tag for child directed treatment=;tfua=;gdpr=$%7BGDPR%7D;gdpr consent=$%7BGDPR CONSENT 755%7D;ltd=;dc tdv=1 —Get your tokens’ worth with Jira In this episode, we cover: 00:00 https://www.youtube.com/watch?v=dChiPsd0uF4 Intro and what Muse Wonder is 01:19 https://www.youtube.com/watch?v=dChiPsd0uF4&t=79s Meta’s frontier capability claims 04:36 https://www.youtube.com/watch?v=dChiPsd0uF4&t=276s Muse code 06:06 https://www.youtube.com/watch?v=dChiPsd0uF4&t=366s Benchmark overview and how the How I AI bench works 07:51 https://www.youtube.com/watch?v=dChiPsd0uF4&t=471s Overall results and TL;DR verdict 11:53 https://www.youtube.com/watch?v=dChiPsd0uF4&t=713s Agentic tasks 13:36 https://www.youtube.com/watch?v=dChiPsd0uF4&t=816s&pp=0gcJCWMAwfN6Pr3D Email writing and voice assistant performance 14:35 https://www.youtube.com/watch?v=dChiPsd0uF4&t=875s Where Muse falls short 18:28 https://www.youtube.com/watch?v=dChiPsd0uF4&t=1108s SVG illustration results 19:33 https://www.youtube.com/watch?v=dChiPsd0uF4&t=1173s Short-form video editing 21:18 https://www.youtube.com/watch?v=dChiPsd0uF4&t=1278s Fast on one-shot work, slow on agent tasks 22:15 https://www.youtube.com/watch?v=dChiPsd0uF4&t=1335s Final verdict and where I’ll use Muse Wonder next Tools referenced: • Meta Muse Wonder / Muse Wonder Full: https://research.meta.ai/blog/muse-wonder https://research.meta.ai/blog/muse-wonder • Muse Code: https://developer.meta.com/ai/lp/muse-code https://developer.meta.com/ai/lp/muse-code Where to find Claire Vo: ChatPRD: https://www.chatprd.ai/ https://www.chatprd.ai/ Website: https://clairevo.com/ https://clairevo.com/ LinkedIn: https://www.linkedin.com/in/clairevo/ https://www.linkedin.com/in/clairevo/ Production and marketing by https://penname.co/ https://penname.co/ . For inquiries about sponsoring the podcast, email \ email protected\ https://www.lennysnewsletter.com/cdn-cgi/l/email-protection .