What Happened
TypeSafe AI’s early-access Jev model is being tested as a fast decision layer for software, including a public Doom demonstration reviewed by Franklin AI.
What Happened
Franklin AI connected Jev to Doom and used structured game state to request actions such as movement and combat decisions. The demo shows an agent acting inside the game loop; it does not show the model reading raw screen pixels.
TypeSafe AI says Jev returns predefined typed values with confidence scores rather than free-form text, with claimed response times of 70–500 milliseconds. The official pricing page charges for input tokens and describes output tokens as free, correcting the broader “zero-cost” wording in the original draft.
Why It Matters
Real-time games need predictable outputs and tight latency budgets. A decision model with a bounded output space may be easier for developers to test and constrain than a general chatbot, but the current performance and reliability figures come from the vendor and need independent production evidence.
Original Source
Watch Franklin AI’s Jev review
