Diogo Almeida, a co-inventor of ChatGPT, has emerged from stealth with TypeSafe AI and its new Jev model. This isn’t another chatbot. Jev is a “System One Model” built to handle deterministic logic directly within production codebases. It bypasses the slow, expensive process of generating text tokens. Instead, it takes unstructured state inputs and outputs type-safe structured values in a single parallel query. The goal is simple: eliminate hallucinations and syntactic failures by design.
The architecture relies on Reinforcement Learning for Calibrated Decisions (RLCD). Unlike traditional models trained on human feedback, Jev focuses on execution logic accuracy. Its hardware-aware parallel sampler evaluates up to 255 discrete options simultaneously. This approach removes the need for complex parsing pipelines or guardrail wrappers. Internal tests show end-to-end latencies between 70 and 500 milliseconds. That is a stark contrast to the 3–329 seconds typical of conversational frontier models. In specific workflow evaluations, Jev proved up to 193.6 times faster than competitors like GPT-6 Astra.
Cost efficiency is equally dramatic. Input processing runs at $0.042 per million tokens, significantly below the $0.20 to $10 range for standard conversational AI. TypeSafe does not charge for output tokens since the model generates structured states without autoregressive passes. Real-world stress tests highlight this advantage. A reactive bot playing Doom operated at 10 queries per second for roughly $7 an hour. Another test navigating Wikipedia links finished with fewer steps and zero hallucinated dead ends.
This shift suggests a bifurcation in the AI market. Conversational models will remain dominant for creative or open-ended tasks. But for high-speed, rule-based automation, probabilistic text generation is becoming obsolete. Jev targets real-time feature extraction and petabyte-scale data workflows where hand-written rules often break. With early developer access now open, engineering teams can finally integrate AI logic without sacrificing speed or budget. The era of using sledgehammers to crack nuts may be ending.
