Leaked Meta benchmark page shows internal "Muse Wonder" beating GPT-6 and Claude Opus 5.5
rohanpaul_ai · x · 2026-10-13
rohanpaulai shared a screenshot of what appears to be a Meta "Frontier Capabilities" page comparing unreleased frontier models on three in-house benchmarks.
- Personal Superintelligence Index: Muse Wonder Full leads at 59.1%, ahead of Claude Opus 5.5 (55.2%) and GPT-6 Astra (max) (54.6%)
- AutomationBench-AA: Muse Wonder Full scores 45.4%, vs 41.2% for Claude Opus 5.5 and 32.1% for Claude Fable 5.1
- MemoryBench: Muse Wonder variants top at 85.6%, above GPT-6's 81.4%
The chart references unreleased models like GPT-6, Claude Opus 5.5 and Claude Fable 5.1. Authenticity is unverified — it could be an internal page leak, placeholder, or speculative content.
Related event: Meta's Muse Models Draw Wave of Early Praise as Execs Amplify Endorsements(9 posts)→
More from Models
- Grok is 'absolutely unhinged' in group chats, user reports — Kyrannio · 2026-10-13
- Power user: Fable is 'too smart for its own good' at everyday knowledge work — evielync · 2026-10-13
- Leaked teaser points to Gemini 4 Argon launch within hours — gaganghotra_ · 2026-10-13
- User asks Google exec about Gemini 4 at Dolores Park, gets ignored; "Argon" codename surfaces — apples_jimmy · 2026-10-13
- Musk amplifies Grok Bot 'Memory Transfer' tool that imports your ChatGPT/Claude memories — elonmusk · 2026-10-13
- Why China may refuse to release open models beyond a certain capability threshold — abhiadesai · 2026-10-13