NUS Introduces Jeva: A Native Architecture for Decision-Centric LLMs

1 reports archived1 independent sourcesupdated 10/26/2026, 12:10:00

Synthesis & Latest Updates

1 Sources Cross-Validated

The NUS xML Lab has introduced Jeva, a novel large model architecture designed to directly output choices, probabilities, or scores. By embedding decision principles like 'null-awareness' and 'order invariance' into its structure, Jeva addresses the inefficiencies and logical flaws of traditional generative models in classification and scoring tasks. The Jeva-4B model and code are now open-source.

LATEST/The NUS xML Lab has introduced Jeva, a novel large model architecture designed to directly output choices, probabilities, or scores. By embedding decision principles like 'null-awareness' and 'order invariance' into its structure, Jeva addresses the inefficiencies and logical flaws of traditional generative models in classification and scoring tasks. The Jeva-4B model and code are now open-source.

Hourly heat curve

3 fully observed hours

Now

9.1

Peak

9.622:00

24h change

–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

Coverage timeline

Total 1 reports · Latest first

新智元 (微信公众号)T2Oct 26, 12:10·78 pts
NUS Introduces Jeva: A Native Architecture for Decision-Centric LLMs
- Jeva architecture bypasses long reasoning chains to directly output structured decisions (choices/probabilities/scores).
- Embeds four structural properties, including Null-awareness (identifying no-correct-answer scenarios) and order invariance.
- Provides Jeva-4B model weights and GitHub repository for developer reproduction and testing.
View Related Reports & Analysis Original Source