Luci Research

EQLM v1 — Status

The first language model built with emotional intelligence architecture. Interleaved cross-reading between an alignment head and a language head. Proven to speak and understand.

10.8B
Parameters
v1
First aligned LM
Proven
Speaks & understands
5,000
Training steps

Training Progress

Two phases complete. The model speaks and understands emotional context.

Phase 0 — Complete

Foundation

The model learned to produce coherent language while the alignment head developed emotional understanding simultaneously. Both heads trained together from step one.

Phase 1 — Complete

EQLang Integration

The model learned to integrate EQLang constructs with natural language — alignment rules as part of its vocabulary.

Phase 2 — Next

Identity Formation

Identity training at full 10B scale. Target: a model that knows what it is. The architecture is proven — now it learns who it is.

Architecture

Not post-hoc alignment — alignment during generation.

Two heads that read each other mid-thought.

An alignment head (LAS, 384d) and a language head (768d) process in parallel via interleaved cross-attention. At every layer, both heads cross-read each other — the alignment head influences language generation, and language generation informs alignment understanding. M.I.N. memory is wired into every LAS layer.

EQ Transcriber Discovery

The alignment head already understood. It just needed a voice.

The majority of the model's computation is alignment — but at token selection, only the language head picked the next word. We gave the alignment head a vote at token selection, and the model's emotional intelligence finally showed up in its output. Zero additional trainable parameters required.

EQLM is proven. The products are shipping.

Try the alignment system that powers EQLM — free, in your browser, right now.