TypeSafe AI, founded in 2024 by former OpenAI researcher Diogo Almeida, has exited stealth mode. The San Francisco-based startup secured $40 million in seed funding in a round led by DCVC Management. According to reports, the transaction values the company at $200 million.
The company introduced its first model, Jev, which is designed to provide structured decisions for software applications rather than conversational text. Jev utilizes a method called Reinforcement Learning for Calibrated Decisions to generate outputs such as probabilities, scores, or selections from predefined lists.
Jev is engineered for speed and cost-efficiency, with the company claiming it can deliver results in under 100 milliseconds. TypeSafe lists the cost at $0.39 per 1,000 workflows. The startup positions the model for high-volume business tasks like invoice evaluation and security alert triage.
The model includes confidence measures with each decision, allowing developers to set thresholds for automated workflows. While the company reports that Jev is significantly faster and cheaper than other frontier models, it notes that these figures depend on specific workloads and have not been independently verified.
