Startup TypeSafe AI, co-founded by former OpenAI researcher Diogo Almeida, has introduced Jev, an AI model designed to make rapid classification judgments within software rather than generating text responses. Almeida previously co-authored the InstructGPT research that laid the foundation for ChatGPT.
Rather than outputting conversational text or code, Jev functions by scoring predefined options and returning probabilities to host software. TypeSafe claims the model delivers responses in 70 to 500 milliseconds—significantly faster than standard language models—by skipping token generation and evaluating multiple output questions in parallel at a cost of $0.042 per million input tokens with no charge for outputs.
While Jev eliminates conversational hallucinations by constraining outputs to defined schemas, its current evaluations rely on internal company benchmarks rather than independent validation. The startup faces competition from existing LLM providers offering structured output features, making real-world speed and accuracy critical to its adoption.
Why it matters
Sub-second latency: Latencies of 70-500ms enable pre-execution validation and real-time routing for enterprise software workflows.
Ultra-low costs: Pricing at $0.042 per million input tokens with free outputs dramatically lowers operational overhead for deterministic classification.
OpenAI competition: Competes with native structured output tools from major LLM providers, requiring superior speed and cost performance to capture market share.
Source: the-decoder.com
