Alexandr Wang, Meta's chief AI officer, describes Muse as a personal AI agent built in seven months and articulates an escalator of agency vision
Meta Launches Muse Wonder, Claims AI Frontier Parity With Rivals
Meta is claiming parity with the two labs considered frontier leaders, backed by benchmark results that are partly confirmed by independent testing and partly contested by price-adjusted comparisons. The API restriction on the stronger model limits developer access to the performance Meta is advertising.
The full picture
Meta released Muse Wonder and Muse Wonder Full on October 12, 2026, with Mark Zuckerberg telling the Financial Times that Meta has caught up with OpenAI and Anthropic at the AI frontier. The models are Meta's largest and most capable yet, designed to handle multistep agentic tasks in personal and professional settings, including booking reservations and managing professional workflows without stopping to ask users for help. Muse Wonder powers the subscription tier of Meta's personal AI assistant called Muse, available at $20 and $100 monthly tiers.
On Meta's own benchmarks, Muse Wonder Full scores 59.1 on the Personal Superintelligence Index, ahead of Claude Opus 5.5 at 55.2 and GPT-6 Astra at 54.5. On SWE-Marathon, Muse Wonder scores 57.5 versus 46.3 for Claude Opus 5.5. On DeepSWE 1.1, with Meta citing independent third-party runs, Muse Wonder Full scores 79.0 against 74.2 for Claude Opus 5.5. On the Harvey legal agent benchmark, which penalizes hallucinations, Muse Wonder Full scores 13.1 versus 4.2 for Claude Opus 5.5. Muse Wonder Full wins 7 of 14 rows on Meta's own benchmark chart against Claude Fable 5.1, GPT-6 Astra, and Claude Opus 5.5.
Independent testing by Artificial Analysis places Muse Wonder Full at 56 on its Intelligence Index, tying Claude Sonnet 5.5 and outperforming GPT-6 Astra, Gemini 4 Argon, and Claude Fable 5.1, each at 53. Claude Opus 5.5 at 58 ranks higher on the independent index. The standard Muse Wonder scores 52, one point below GPT-6 Astra. At the same per-token price, Claude Opus 5.5 scores six points higher on the independent index and wins 9 of 14 rows on Meta's own chart. Claude Sonnet 5.5 matches Muse Wonder Full's independent score at half the price.
Muse Wonder Full is not available via API; only the weaker standard Muse Wonder is accessible through the API. Meta's MemoryBench scores Wonder at 85.6 and Wonder Full at 85.5, tested on long conversations of roughly 64,000 to 100,000 tokens, but those results do not confirm reliable recall across the full 1 million token window or across unlimited separate conversations. Zuckerberg said Meta is already working on the next step up and brushed off calls to slow down.
Alexandr Wang, Meta's chief AI officer, described Muse as a personal AI agent built in seven months and articulated a vision of AI as an escalator of agency that preserves the ambitions people typically lose as they enter the workforce.
How it developed
Meta releases Muse Wonder and Muse Wonder Full; Zuckerberg tells Financial Times Meta has reached AI frontier parity with OpenAI and Anthropic
Sources
3 more sources
Related
- Split fromClaude 5.5 family and GPT-6 Sol
Want this in your inbox?
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free