Scaled Cognition, an AI lab prioritizing reliability over raw capability, has secured $100 million in Series A funding. The round was led by Khosla Ventures at a $750 million valuation, according to the Wall Street Journal.
The company develops the Agentic Pretrained Transformer model, an architecture designed for stability in deployed AI systems. Its focus on dependability stands in contrast to many labs racing for scale and benchmark performance.
The CEO characterized foundation models as “like schizophrenic geniuses,” highlighting the industry's reliability gap. The round suggests investors are betting on specialized, trustworthy AI over general-purpose models.
The firm now faces the challenge of proving its technology at scale while competing against well-funded rivals. Adoption by enterprise clients will be a key test of its reliability-first thesis.
Some analysts caution that prioritizing reliability may constrain capability, potentially limiting the model's utility for complex, open-ended tasks. The balance between safety and performance remains unresolved.