[KongchangAI]
Model

Jav

A compact on-device language model developed by TypeSafe AI, designed for structured output and function calling. It occupies the same lightweight tool-calling model category as Neural 3, targeting deployment in resource-constrained edge environments.

Timeline (last 90 days)

Oct 5

Jav is launched by TypeSafe AI (founder Diego Almeida), produces the entire structured answer in a single parallel pass, has a response time of 70 to 500 milliseconds, and is a large cloud-based model.

Unverified50%
Oct 5

The Register pointed out that Jav's claimed 'no hallucination' actually only means it cannot produce an illegal structure, and it can still select the wrong option.

Unverified50%

All Facts (2)

Source Articles