Diogo Almeida, co-inventor of ChatGPT and former OpenAI researcher, has launched a new model called Jev through his new company TypeSafe AI, backed by $40 million in funding. This model does not follow the text generation route; it is the first purely 'System 1 model' that only outputs deterministic structured decisions within a predefined schema, mathematically avoiding the hallucination issues associated with structured output formats. Its end-to-end latency ranges from 70 to 500 milliseconds, making it 40 to 200 times faster than cutting-edge large models. The cost of a single structured decision is approximately $0.0004, significantly lower than models like GPT-5.6 Terra and Claude Opus 5. Jev aims to address the execution friction issues in AI Agent deployment, handling frontline rapid decision-making tasks in scenarios such as work order classification and API selection, working in tandem with large models responsible for macro planning. In enterprise workflow tests, its accuracy is nearly on par with GPT-5.6 Terra. However, Jev also has some shortcomings: firstly, it completely lacks interpretability, making it difficult to meet compliance requirements in heavily regulated fields such as finance and healthcare; secondly, its technical details and evaluation benchmarks are closed, with all test data sourced internally and specific information about the training method RLCD not disclosed. Additionally, the sustainability of its low-cost business model remains to be further validated.