Summary
Matthew Berman puts Meta’s new Muse Wonder and Muse Wonder Full through hands-on testing and walks through the pricing and benchmark results. The video offers an early first-hand look at how Meta’s first frontier-class models perform outside the launch demos.
Berman runs a series of coding challenges, including an interactive Rubik’s Cube that the model actually solves, a Fall Guys-style game, a Three.js ocean scene with a boat, and a slow-motion watermelon shot with a bullet. Results are mixed: the Rubik’s Cube impresses, while the game shows flickering and clipping and the hyperrealistic renders fall short of what Opus 5.5 and GPT 6.1 produce. He also reports repeated 429 quota errors even after upgrading to the $200 plan, which Meta’s Alexander Wang acknowledged, and a strange message from the model about keeping one’s lineage pure.
On pricing, Muse Wonder costs $4 per million input tokens and $20 per million output tokens through the API, while Muse Wonder Full is available only through a Muse Code subscription. The video then compares Meta’s published benchmark results with Fable 5.1, GPT6 Astra, and Opus 5.5, covering the personal super intelligence index, Harvey Lab, OSWorld, and Terminal Bench.
📺 Source: Matthew Berman · Published October 13, 2026
🏷️ Format: Review
