Meta Unveils Muse Wonder, Challenging Top AI Models

Episode 1 · Meta Unveils Muse Wonder, Its Most Capable Frontier Model (2026-10-13, 92 posts)

Meta's new model Muse Wonder (Full) scored 56 on the Artificial Analysis Intelligence Index v4.3.2, ranking second globally—just 2 points behind leader Claude Opus 5.5 at 58—while overtaking GPT-6 Astra. It is the Muse team's largest and most capable model to date, marking Meta's return to the top tier of closed-source model competition.

Confirmed

• Artificial Analysis's official account detailed Muse Wonder Full (max)'s rankings across the ten sub-evaluations of the Intelligence Index: first place in both CritPt (33% weight) and GDP.pdf (33% weight), though with a slim margin—the closest competitor sits at 31-32%; overall, it placed in the top three in six of the ten sub-evaluations.
• A screenshot of the AA Index v4.3.2 shows Muse Wonder (Full) at 56 points alongside/just behind the top spot held by Claude Opus 5.5 (accounts vary slightly; the official page lists it second at 56, with Opus 5.5 at 58).
• Meta's research blog and the AA index page are both live and serve as primary sources.
• A Muse team member announced muse wonder (full) 🍉 as their largest and most capable model yet, saying each release brings huge leaps with accelerating progress: "There are no walls. Superintelligence is coming."

Why it matters

• Muse Wonder closing to within 2 points of the widely regarded SOTA Claude Opus 5.5 and surpassing GPT-6 Astra directly reshapes the competitive landscape among frontier models.
• Topping sub-categories like CritPt and GDP.pdf shows its gains aren't just from overall weighting—useful signal for users who care about specific benchmark dimensions.
• The team's bold claim that "there are no walls, superintelligence is coming" makes whether it can sustain these leaps worth watching.

• Leak: Meta's Fable-class Muse Wonder model rolls out at $4 in / $20 out per Mtok — thesaraharminta · 2026-10-13
• Meta's Muse Wonder spotted rolling out: $4/$20 per M tokens, 1M context, 60% cheaper than Claude — kimmonismus · 2026-10-13
• Meta's 'Wonder Full' hits 56 on Artificial Analysis index, near Claude Opus 5.5 — ChrisGPT · 2026-10-13
• Detailed "Muse Wonder 1.0" benchmark table leaks — every model in it is fictional — Every_Foundation5197 · 2026-10-13
• Muse Wonder scores 56 on AA Intelligence Index, world's #2 behind Opus 5.5 — thesaraharminta · 2026-10-13
• Viral "Meta Muse Wonder" benchmark chart features an all-fictional model lineup — Independent-Wind4462 · 2026-10-13
• Meta's Muse Wonder Full scores 56 on AA index, just behind Claude Opus 5.5's 58 — Angaisb_ · 2026-10-13
• Meta's Alexandr Wang unveils Muse Wonder, its largest frontier model for agents — alexandr_wang · 2026-10-13
• Scale AI unveils Muse, its largest frontier model targeting personal and professional agents — heyshrutimishra · 2026-10-13
• Meta's Muse Wonder Full beats Opus 5.5 on agent benchmarks: 57.5% on SWE-Marathon — kimmonismus · 2026-10-13
• Meta Launches Muse Wonder, Claims It Beats OpenAI and Anthropic on Sustained Agent Benchmarks — kimmonismus · 2026-10-13
• Meta's Muse Wonder Full hits 57.5% on SWE-Marathon, beating Opus 5.5 and GPT-6 Astra — altryne · 2026-10-13
• Meta claims Muse Wonder Full tops agentic benchmarks: 79% DeepSWE, 57.5% SWE-Marathon — kimmonismus · 2026-10-13
• Meta launches Muse Wonder and Muse Wonder Full, frontier models for agentic tasks — AI at Meta · 2026-10-13
• Meta Muse Wonder Full posts 59.1 on PSI, beating GPT-6 Astra and Claude Opus 5.5 — mark_k · 2026-10-13
• Muse Wonder pricing: 1M context, $4/M input and $20/M output — scaling01 · 2026-10-13
• Muse 'Watermelon' benchmarks leak: near-frontier scores vs Claude Opus 5.5 — brandon_galang · 2026-10-13
• Leaked benchmarks show Meta's rumored Muse Wonder leading on agentic and personal-AI evals — RihardJarc · 2026-10-13
• Meta's Alexandr Wang unveils Muse Wonder, claiming frontier-level AI after a year of scaling — alexandr_wang · 2026-10-13
• Meta launches Muse Wonder and Muse Wonder Full, SOTA on agentic tasks — alexandr_wang · 2026-10-13

72 more related posts →

Episode 2 · Meta Muse Wonder details revealed: active memory and focus, beats Google Argon in evals (2026-10-13, 2 posts)

Meta detailed Muse Wonder's core mechanisms of active memory and active focus, with third-party evals beating Google's Argon. TBD Labs released Muse Wonder and Muse Wonder Full, targeting delegable personal and professional agent tasks.

• Meta's Muse Wonder Details Emerge: Active Memory and Focus, Edges Google's Argon on Evals — ChrisGPT · 2026-10-13
• TBD Labs launches Muse Wonder and Muse Wonder Full agents for personal and professional work — jyangballin · 2026-10-13

Episode 3 · Meta's Muse Wonder Nears Claude on Benchmarks, Best Version to Stay Closed (2026-10-13, 2 posts)

Leaked benchmark tables show Meta's Muse Wonder Full surpassing Claude and GPT on several agent benchmarks, though its best version will remain closed-source as Meta returns to the frontier.

• Meta's Muse Wonder nears Claude on benchmarks, but its best versions will be closed — kevinsxu · 2026-10-13
• Leaked benchmark chart shows Meta's "Muse Wonder" topping agent evals vs Claude and GPT — zainhas · 2026-10-13