Show HN: Using Jev to generate game levels in real time (www.spritefusion.com)

TypeSafe has unveiled Jev, a novel model designed to generate game levels in real time with structured outputs. Unlike traditional text-generating models, Jev functions as a zero-shot classifier, producing timely decisions with associated probabilities rather than verbose text. This is a game-changer for the AI/ML community, particularly in the realm of gaming, where the dual challenges of latency and cost have hindered the extensive use of AI. Jev promises sub-second response times at a cost-efficient rate of $0.042 per million input tokens, far more affordable than existing large language models. In practical tests, Jev successfully generated dynamic levels for a runner game, maintaining an average API response time of 319–375 ms and incurring minimal costs of about $0.00057 per request. The model's ability to process real-time game state inputs—like player position and terrain details—while keeping latency manageable opens new avenues for creative development in game design. Jev’s structured output capabilities not only enhance interactivity but also encourage innovative experimentation in procedural level generation, marking a significant step forward for AI-driven gaming solutions.