Samaya AI's Sterling: a 30B MoE model post-trained for financial research claims frontier-level performance
JeffDean · x · 2026-10-30
Samaya AI introduced Sterling, a model post-trained for challenging financial research that the company says outperforms frontier models. It began as a two-person effort to improve agentic orchestration efficiency and ended up post-training a much smaller model — Nemotron 3.5 Lightning, a 30B MoE with 3B active parameters — to greater accuracy. Jeff Dean congratulated the team.
More from Venture
- Sentry CEO pushes back: software businesses don't need $8M+ seed rounds to survive — zeeg · 2026-10-30
- Amazon guides AWS to 40%+ growth through 2028, eyes $1T business in 5-10 years — mitchdeg · 2026-10-30
- Anthropic IPO tilts to $1.5T: 2023's $1M grants now worth ~$78M — FinanceYF5 · 2026-10-30
- Anthropic IPO seen closer to $1.5T than $2T, early employee stakes swing 52x — FinanceYF5 · 2026-10-30
- Chart: a $1M Anthropic grant at 2023's Series C is now worth $78M — FinanceYF5 · 2026-10-30
- Matthew Berman: Anthropic Likely Won't Release New Models Until After IPO — MatthewBerman · 2026-10-30