Zuckerberg admits he wasn't sure Meta's $200B frontier model bet would work
MaxUnfried · x · 2026-10-14
A widely shared tweet quotes Zuckerberg saying that when Meta decided to write a $100–200 billion check for its first frontier model (called Muse Wonder), he asked Alexandr Wang "are you sure this is going to work?" — and Wang replied it was "certainly not clear at all that it was going to work." The poster's takeaway: the biggest decisions are made on vibes — executives review the numbers, then make a gut call.
The tweet also lists unverified claims: Meta's internal tests put the model level with Fable 5.1 and GPT-6 Astra; it beats Fable, Opus, and Astra on MemoryBench; priced at $4 input / $20 output per million tokens; trained at Prometheus, Meta's new Ohio data center; no independent benchmarks yet.
Related event: Zuckerberg: committed billions to AI before knowing it would work(2 posts)→
More from Companies & People
- Hugging Face's first investor Matt Hartman on why he only backs technical founders who can pivot — every · 2026-10-15
- SpaceXAI vs Google: analysts peg third-place AI player's run rate at $4.5–5B alongside Gemini — thedealdirector · 2026-10-15
- Mirror Mirror AI books real models' likenesses for AI fashion campaigns at 90% lower cost — _jasonwei · 2026-10-15
- Bezos says AI productivity could mean 3-day work weeks, pressed on Amazon layoffs live on air — VraserX · 2026-10-15
- Marty Cagan: the real trade-off for product teams is outcomes vs predictability, not output — rseroter · 2026-10-15
- Vercel dev-rel lead departs after nearly two years, built Startup Program and AI Accelerator — paw_lean · 2026-10-15