Event · User reviews

Meta's Muse Wonder: what people who tried it said

Hands-on reviews posted on X in the first day. Posts from Meta's official accounts and staff are left out, and both praise and criticism are included.

1. Positive@tobi · X ↗

   'Struggled to find anything it couldn't do'

   Shopify CEO Tobi Lütke, who had early access, said it handled tasks other top models failed and that 'Meta is back.' The post got 3,700 likes.

2. Positive@theo · X ↗

   Don't judge it by 3D game demos

   Developer YouTuber Theo said 3D demos don't represent real coding work and called Wonder an incredible code model.

3. Positive@_MaxBlade · X ↗

   Same prompt, a twentieth of the cost

   Asked to build a lawn mower simulator, the API-equivalent cost was $4.11 for Muse Wonder and $99.43 for Opus 5.5, the user posted.

4. Mixed@emollick · X ↗

   Seems good from limited testing

   Wharton professor Ethan Mollick said he had early access but little time, and it seemed like a good model in limited tests.

5. Negative@hakmgpt · X ↗

   A disappointing Minecraft clone

   A user posted a weak result from asking for a Minecraft clone at the highest reasoning setting. Meta later said it hadn't trained on 3D demos.

6. Negative@LLMJunky · X ↗

   Worse than Chinese models on a personal test

   On the user's own RocketLeagueBench, it did worse than nearly every Chinese model they had run.

7. Negative@bijanbowen · X ↗

   Two runs blew past the 5-hour limit

   On the $50 plan, two Muse Wonder Full runs used 108% of the 5-hour limit and 36% of the weekly limit. Meta later fixed a usage-counting bug.

8. Negative@Lentils80 · X ↗

   'Benchmaxxed'

   After seeing many outputs, the user called it a model tuned for benchmark scores, worse in that respect than Gemini.