We’re Scaled Cognition, developing the first ever models trained specifically for agentic applications:. 1. Our first system, APT-1, is now #1 on agentic benchmarks. 2. It was developed by a US team for a total cost of less than $11M. 3. Khosla Ventures led our seed round ($21M

This @FortuneMagazine article outlines the risk of using general purpose LLMs for CX. Leading AI startup @cursor_ai CX agent hallucinated a key policy causing confusion amongst its users. General purpose LLMs lack determinism, and are prone to consequential errors. AI agents,.

Our CTO Dan Klein was featured in @nytimes, sharing insights on building AI agents that can reason through tasks. Accelerate your company’s development of AI agents with APT-1 with its built-in reasoning, action-taking, and decision making capabilities.

We’ve accomplished these gains through (1) optimizing the model for actions rather than tokens, (2) a new kind of synthetic agentic training data, and (3) a novel RL approach using agent-to-agent self play. (1) Standard models are focused on token sequences, but business logic.

APT-1 currently outperforms all other models on the Tau-Bench and ComplexFuncBench agentic leaderboards, which test the ability to invoke sequences of complex APIs and comply with business policies.

We’re also announcing our Agent Builder platform, which allows you to create, test, and deploy an enterprise-grade AI agent using APT-1 in under an hour. Our GenAPI technology lets you test agent behaviors without needing to integrate with real APIs during development. Learn