MOTHER Model Family

One sovereign
brain.

A single 7B world-model family that perceives, reasons and predicts - vision, language, speech and action fused into one sovereign core. Owned, hosted and trained in Britain, on British infrastructure.

100%100%85%100%94%100%
7B
Parameters
48L · 3072d
CORE transformer
26
Trained heads
100%
Sovereign

CORE 7B backbone

A 48-layer / 3072-dim reasoning transformer shared across every product - the one brain behind text, vision and action.

MOTHER DeepVision

A sovereign ViT (27L · 1152d · 896px) driving detection, tracking, faces, expression, spatial relations and scene graphs.

Speech & audio

Voice, speech and audio-perception heads for real-time understanding and generation across the stack.

MOTHER memory

A three-tier agent memory - episodic, semantic and procedural - so the model carries context across sessions.

World model

Latent vision + action dynamics predict what happens next - the substrate for MOTHER EXO and robotics.

Open-weight & auditable

Published under EU AI Act Article 53 with a copyright policy and training-content summary. No black boxes.

Build on MOTHER AI

Access MOTHER CORE, LLM 7B, Quantum RAG and the T2V engine via a sovereign REST API - free tier and enterprise plans.

A built-in agent fleet that deploys with the model

MOTHER ships with a fleet of governed agents - diagnosis, repair, data-curation and a self-evolving world-model loop - that observe, propose and stage improvements. Nothing auto-applies: every change is red-line gated, sim-tested and human-promoted.

  • diagnosis · observes health
  • repair · drafts fixes, never auto-applies
  • self_improve · LoRA in sim, staged as candidate
  • sophia · self-evolving world-model loop
  • data_curator · long-tail coverage
  • Guardian-signed promotion only

Sovereign by construction

The whole family runs on sovereign GB10 / DGX nodes on MSAI infrastructure. Weights and data never leave your control - on-prem by default, with a full model card, checkpoint lineage and reproducible training steps.

Modular cognitive architecture

MOTHER CORE is a British class of LLM with cognition separated into specialized, independently upgradeable domains - deterministic reasoning, a public language engine, and sovereign multi-modal generation, all on one sovereign core.

  • CORE 7B - deterministic reasoning (T=0)
  • LLM 7B - public language engine
  • T2V - sovereign multi-modal (text-to-video)
  • CORE 70B - 4× H200 tensor-parallel (planned)

CORE 7B specializations

Domain experts fine-tuned on the deterministic CORE backbone - each a sovereign reasoning specialist tuned for its field while sharing the same auditable core.

  • CORE-Reasoning - theorem proving
  • CORE-Science - physics simulations
  • CORE-Defence - threat assessment
  • CORE-Legal - compliance
  • CORE-Maths - optimization

LLM 7B specializations

Public language-engine variants tuned for expressive, human-facing work - from narrative and code to policy and teaching, all on the same sovereign core.

  • LLM-Creative - narrative generation
  • LLM-Coder - code & docs
  • LLM-Policy - government writing
  • LLM-Education - teaching
Spec sheet
backbone
CORE 7B
layers
48
width
3072d
vision
DeepVision ViT
precision
bf16
heads
26 trained
license
MSAI open-weight
compliance
EU AI Act Art. 53

See inside the brain.

Explore the layered world-model, its agents and the sovereign core powering every MSAI product.

Inside MOTHER EXO