TypeSafe AI released Jev, a model that returns typed probabilistic decisions for machines. It avoids text parsing by providing structured outputs. Jev uses a System One architecture called RLCD. It processes queries in parallel for faster responses. Input costs $0.042 per million tokens with free output. TypeSafe claims it is 40x to 200x faster than traditional LLMs. It demonstrated Jev playing Doom using structured game state data. The model targets AI automation, real-time decisions and data classification. TypeSafe says it eliminates hallucinations in tool calls. The name references economist William Stanley Jevons.
What changed
AI models returned natural language text that required parsing and validation.
What it unlocks
Returning typed probabilistic decisions directly usable by other software without text parsing.
- 70ms-500ms response time
- $0.042 per million input tokens
- 40x-200x faster than traditional LLMs
- 238x cheaper than Fable 5.1
Sources