Meta Muse Wonder Tested: Is Meta Finally Frontier?
Meta just released Muse Wonder, its first frontier model, and I put it through my VisualForge benchmark on launch day. I stopped the run after five of my nine tests, so every model here is compared on those same five interactive web pages, each scored in a real browser. In this video: • what Meta actually released: Muse Wonder (API, Muse Code, OpenRouter) and Muse Wonder Full (Muse Code subscribers only) • Meta's 2:32 launch video, with my comments on each claim • Alexandr Wang's statements, and what was reported around the launch • Meta's benchmark table, checked row by row against Opus 5.5, GPT-6 Astra and Fable 5.1 • prices and four ways to use it • my own results: 6.86 out of 10, fourth of 16 models on the same five tests, behind GPT-6 Astra (7.44), Opus 5.5 (7.16) and Sonnet 5.5 (7.12) • cost: the same price per token as Opus 5.5 ($4 in / $20 out per million), but about $2.17 a test against Opus's $5.89, because Opus writes about 4.6× as many tokens. Lower bill, lower score: Wonder scored below Opus on four of the five tests Settings: OpenRouter meta/muse-wonder at maximum effort, to match Meta's charts. The top models on my leaderboard (Opus 5.5, GPT-6 Astra) ran at high, one step below, so Wonder had a little extra thinking time here. One run, launch day. Meta's numbers are labelled as Meta's claims. Costs are estimates from each run's token counts at OpenRouter prices (checked 12 October 2026), cached input at its discounted rate; on one test from my earlier medium-effort Wonder run, the same estimate matched my OpenRouter bill to the cent. Every page Wonder built, with its scores, tokens and cost: https://visualforgebench.com/run/?id=20261012T183411_meta-muse-wonder_a28a All results: https://visualforgebench.com (the site's leaderboard uses its Quality v1 scoring over every page a run made, so its ranking differs slightly from the same-five comparison in this video) Chapters 0:00 Intro 0:27 What Meta released 1:37 Meta's launch video 4:58 Meta's benchmarks, checked 6:09 Against the frontier 7:04 Benchmaxxing and the 3D test 8:00 Alexandr Wang's claims 8:51 How to use it 9:46 My benchmark 10:22 My results 13:01 Verdict Sources (read 12 October 2026) - Meta announcement and launch video: https://about.fb.com/news/2026/10/introducing-muse-wonder-models/ - Meta research blog: https://research.meta.ai/blog/muse-wonder - Alexandr Wang on X: https://x.com/alexandr_wang/status/2108456337374065097 - CNBC: https://www.cnbc.com/2026/10/12/meta-muse-wonder-ai-model-agent.html - TechCrunch: https://techcrunch.com/2026/10/12/meta-releases-its-first-frontier-model-muse-wonder/ - The Next Web (Bloomberg interview): https://thenextweb.com/news/meta-muse-wonder-frontier-eu - OpenRouter: https://openrouter.ai/meta/muse-wonder Clips from Meta's launch video are used for commentary and review. Narration: my own voice clone (HeyGen).