ChatGPT pioneer launches Jev model for programmatic logic
ainews ·
TypeSafe AI, founded by ChatGPT co-inventor Diogo Almeida, has officially launched its new model named Jev after a two-year stealth development period. Unlike traditional large language models that generate sequential text tokens, Jev is designed specifically for programmatic logic and automated deterministic decisions within production codebases. The system utilizes a hardware-aware parallel sampling architecture to output type-safe structured values directly, thereby eliminating syntactic errors and hallucinations common in autoregressive generation. Training relies on Reinforcement Learning for Calibrated Decisions (RLCD), which ensures that confidence scores correlate accurately with output precision rather than human conversational preferences. Performance benchmarks indicate significant advantages over existing frontier models, with execution speeds up to 193.6 times faster and response latencies ranging from 70 to 500 milliseconds. Cost efficiency is also highlighted, with input processing priced at $0.042 per million tokens and no charges for output tokens due to the non-autoregressive nature of the model. Real-world tests demonstrate Jev's capability in high-speed game states and complex web traversal tasks, targeting applications in real-time feature extraction and petabyte-scale data workflows. Early developer access is now open as TypeSafe begins onboarding engineering teams from its deployment waitlist.
ainews