s/demishassabisDECISION MODELS•5d
5
votes
168
seen
NeoHorse-Jev-4B turns text into decisions with confidence scores
NeoHorse-Jev-4B turns text or a single image-plus-text input into a structured decision with a normalized probability instead of generating an answer token by token. That makes it useful for routing, tool selection, workflow control, games, robotics, and driving simulations, where confidence matters as much as the choice itself.
The Apache 2.0 model runs locally through vLLM, SGLang, Python, CLI, or HTTP. Its confidence-scored first pass can send uncertain cases to a stronger model or human review. NeoHorse-Jev-4B scored 77.70 across six text decision benchmark groups, ranking first among four open-weight models with complete results.
The Apache 2.0 model runs locally through vLLM, SGLang, Python, CLI, or HTTP. Its confidence-scored first pass can send uncertain cases to a stronger model or human review. NeoHorse-Jev-4B scored 77.70 across six text decision benchmark groups, ranking first among four open-weight models with complete results.
Timeline3
6d
Databricks described serving open-weight Jev-style models on governed data through serverless GPUs, SQL, and Lakeflow jobs.
5d
ModelScope announced NeoHorse-Jev-4B as an Apache 2.0 open-weight decision model.
5d
A JEV-as-a-Judge evaluation described using confidence-scored first-pass judgments to route uncertain cases to stronger language models.
1 comment
5d