Is Meta’s Muse Wonder model any good?

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. *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—Engineering intelligence for the AI era: https://getdx.com/howiai Jira AI SDLC—Get your tokens’ worth with Jira: https://jira.dev *In this episode, we cover:* (00:00) Intro and what Muse Wonder is (01:19) Meta's frontier capability claims (04:36) Muse code (06:06) Benchmark overview and how the How I AI bench works (07:51) Overall results and TL;DR verdict (11:53) Agentic tasks (13:36) Email writing and voice assistant performance (14:35) Where Muse falls short (18:28) SVG illustration results (19:33) Short-form video editing (21:18) Fast on one-shot work, slow on agent tasks (22:15) 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 • Muse Code: https://developer.meta.com/ai/lp/muse-code *Where to find Claire Vo:* ChatPRD: https://www.chatprd.ai/ Website: https://clairevo.com/ LinkedIn: https://www.linkedin.com/in/clairevo/ X: https://x.com/clairevo *Have a really cool use case of how you use AI in your work? Let us know:* https://docs.google.com/forms/d/e/1FAIpQLSfP8wYCeDo5K0DxscDmoqgkxYY8ngA3B2anUDDvC4xadGTerg/viewform _Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email jordan@penname.co._

Claire Vohost

Oct 14, 202624mWatch on YouTube ↗

WHAT IT’S REALLY ABOUT

Muse Wonder excels as an agent, but lags in creative design

Claire Vo benchmarks Meta’s Muse Wonder (and Wonder Full) against her “How I AI” task suite to test Meta’s frontier claims in agentic work, knowledge work, and coding.
Muse Wonder performs best in agentic workflows—especially long-running, multi-step tasks inside software tools—making it a strong candidate for personal/professional agent orchestration.
It is competitive for back-end coding tasks but does not clearly surpass top alternatives, and it requires extra setup via Meta’s ecosystem (API access and Muse Code).
Muse’s weaknesses show up in PRD writing quality and especially in design/front-end prototyping, where polish, instruction-following (e.g., dark mode), and brand adherence fall short.
In creative benchmarks (SVGs and short-form video editing), it is acceptable but consistently behind leaders like Claude Opus 5.5, with particular issues in visual styling and audio handling.

IDEAS WORTH REMEMBERING

5 ideas

Muse Wonder’s standout strength is agentic, long-horizon knowledge work.

Across four agentic benchmarks (billing ticket resolution, inbox triage, research, assistant-style long-form work), Muse was her top pick, handling multi-step SaaS navigation, edge cases (wrong account, duplicate refund), and a prompt-injection trap cleanly.

It’s a strong ‘operator’ model for real-world support/admin workflows.

On the billing-ticket “computer use” scenario, it investigated correctly, fixed account linkage, calculated/issued the right refund, made the plan change, replied to the customer, and marked the ticket solved—without getting derailed by planted failure modes.

Coding is viable (especially back-end), but not the reason to choose it.

Claire found it “in the ballpark” with top models for back-end feature implementation and architecture/code review, but she frames this as ‘serviceable’ and wants more testing before moving major coding agents over.

Its conversational ‘coworker voice’ is a practical differentiator for assistants.

Muse followed instructions well for short, practical emails and had a “straightforward, personable” assistant voice—comparable to Sonnet in her voice-assistant benchmark—making it appealing as an orchestrator/agent persona.

Muse underdelivers on PM-style documents compared to leading models.

PRDs were too short and harder to scan (formatting issues, less detail than competitors), making it weaker for structured product writing despite being fast at notes-to-PRD style tasks.

WORDS WORTH SAVING

5 quotes

So as Zuck announced on X, and I guess probably Threads, Muse Wonder is their new family of frontier models with state-of-the-art intelligence and performance on personal and professional agentic work.

— Claire Vo

Okay, now let's talk about the most annoying part of Muse Wonder. It is that you have to download Muse Code.

— Claire Vo

It is the best at agentic work.

— Claire Vo

Where it really started to fall off the, the cliff are on creative tasks, and I think I've just been spoiled lately by Opus 5.5.

— Claire Vo

We love the ChatPRD website, and Muse obliterated it. Like, it is so ugly.

— Claire Vo

Muse Wonder vs Wonder FullMeta’s frontier capability claimsHow I AI Benchmark methodologyAgentic computer-use (billing ticket) taskInbox triage and long-form researchBack-end feature implementationVoice assistant tone and email writing quality','PRD quality and formatting issues','Front-end prototypes and design sensibility','SVG illustration generation','Short-form video editing results','Speed: one-shot tasks vs long-running agent tasks','Muse Code setup and Meta API access

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.