TypeSafe AI Releases Jev Decision Models for Structured Classification and LLM Judging
Context that changes how you build, even if there's nothing to install.
On September 22, 2026, Simon Willison released the 'llm-typesafe' CLI plugin to support structured classification and ranking using the Jev model.
It is a niche utility for specific structured output tasks with very limited general application.
Useful for fans of Simon's toolset, but a minor footnote in the broader AI ecosystem. Structured outputs are increasingly a native feature of frontier APIs anyway.
Native integration of Jev-like classification in major LLM frameworks.
Also covers
- TypeSafe AI Releases Jev Decision Models for Structured Classification and LLM JudgingTypeSafe AI has released 'Jev,' a suite of decision models designed to solve the 'LLM-as-a-judge' problem. Unlike general models, Jev is optimized for evaluating the accuracy and consistency of responses from other models without the biases typical of general-purpose LLMs.
- Introduces specialized, constraint-focused models for programmatic decision-making.
- Prioritizes predictable structured outputs (like 'noul' types) over traditional conversational fluency.
- Signals a shift toward LLM integration in reliable classification pipelines like support ticket triage.
Sources agree that the llm-typesafe plugin was released by Simon Willison.
The nature and existence of the underlying 'TypeSafe AI' service and 'Jev' model lack mainstream corroboration beyond the developer's blog.