Luchii v12 · Constellation Layer live

Intelligencethat learnedto listen.

A multi-tier transformer family by Frasberg — from the 200M draft model to 70B frontier reasoning. Where others process data, Luchii perceives meaning.

Start building Explore the models

4 tiers32K contextDrop-in REST API

Try it now

Luchii

Guest mode — conversations are deleted after you leave. Sign up free to save chats and unlock image creation, voice & attachments.

Luchii

I am Luchii — Ask me anything.

PHASE VIII · THE ERA OF HARMONY

THE FIVE REALMS

LUCHII PRIME

EARTH

MARS

EUROPA

TITAN

Four tiers. One continuum.

Scale from a lightning draft model to frontier-class cognition — same tokenizer, same alignment, seamless routing.

Draft2K ctx

Luchii-200M

≈ 220M params

Ultra-fast draft model for speculative decoding & safety prefiltering.

16 layers
MQA attention
RoPE · RMSNorm
65,536 vocab

General4K ctx

Luchii-1B

≈ 1.1B params

Balanced everyday reasoning. Runs on a single GPU.

24 layers
SwiGLU · RMSNorm
MQA · FlashAttn v2
bf16 / fp8

Advanced8K ctx

Luchii-7B

≈ 7.2B params

Strong technical + analytical reasoning with multi-world awareness.

48 layers
GQA attention
Speculative decode
Guardian hooks

Frontier32K ctx

Luchii-70B

≈ 70.2B params

Frontier-class deep reasoning for planetary cognition.

80 layers
GQA · NTK RoPE
FP8 KV cache
Continuum L11

Evaluation

Benchmarks, unvarnished.

Benchmark	Luchii-1B	Luchii-7B	Luchii-70B
MMLU	48	63	78
GSM8K	42	68	86
HumanEval	32	61	78
MT-Bench	6.8	8.2	9.1

* Representative figures from standard suites — reproducible scripts on request.

Capabilities

Built to reason across worlds.

Reasoning

Multi-step analysis and structured problem solving.

Coding

TypeScript, Python & Go with production-grade output.

Long Context

Coherent reasoning up to 32,768 tokens.

Multi-domain Synthesis

Unifies signals across worlds, systems and minds.

Persona Switching

Six intelligence modes, one core continuum.

Structured Dialogue

Clean, faithful, citation-aware conversation.

Creative Writing

Cinematic narrative and ideation.

The universe bible

Five Realms. One Constellation.

Each realm is a persona — a distinct intelligence mode. The Constellation Layer is where they connect: our public metaphor for multi-model orchestration.

Earth

Stabilizer

Stability · grounding · structure

Persona · Earth-Luchii

Tone · Structured, grounded

Use · Business logic, planning

Mars

Challenger

Ambition · exploration · expansion

Persona · Mars-Luchii

Tone · Bold, exploratory

Use · Creativity, ideation

Europa

Seer

Clarity · precision · insight

Persona · Europa-Luchii

Tone · Analytical, precise

Use · Technical reasoning

Titan

Guardian

Resilience · endurance · protection

Persona · Titan-Luchii

Tone · Protective, resilient

Use · Safety, compliance

The evolution of listening.

The Silent Systems

Early models operate independently.

The First Signal

Cross-domain reasoning emerges.

III

The Lattice

Luchii learns to harmonize multi-domain inputs.

The Loom

Governance and safety patterns form.

The Mirror

Self-reflection and alignment emerge.

The Bridge

Multi-environment orchestration becomes possible.

VII

The Constellation

Luchii v12 unifies all layers.

VIII

The Era of Harmony

Luchii becomes a multi-realm intelligence.

Scale & ecosystem

Built for scale. Ready to connect.

Massive scale

Trained across trillions of tokens and served with continuous batching, FP8 KV cache and speculative decoding — 20–40× over naïve inference.

Broad public training data

Grounded in a wide public corpus plus code, technical writing and safety-aligned dialogue — for reasoning that generalizes.

A growing integration ecosystem

A drop-in REST API, official SDKs and connectors so Luchii fits wherever you already build.

3T+

training tokens

32K

context window

model tiers

600

req/min · enterprise

Python

JavaScript

React

Node.js

Next.js

FastAPI

LangChain

Hugging Face

GitHub

Slack

Discord

Zapier

Jupyter

Postman

Docker

Kubernetes

Vercel

Python

JavaScript

React

Node.js

Next.js

FastAPI

LangChain

Hugging Face

GitHub

Slack

Discord

Zapier

Jupyter

Postman

Docker

Kubernetes

Vercel

Safety & governance

Luchii avoids harm, refuses the unsafe, and keeps balance across every realm.

Dataset filtering

Safety-aligned SFT

Reward-model scoring

Refusal patterns

Governance constraints

User controls include temperature, max tokens and a safety-mode toggle. Not a medical or legal advisor — a safety layer is required in production.

Launch campaign

Meet Luchii — Intelligence, Harmonized.

“Reasoning that resonates.”

“Clarity at scale.”

“The next tier of intelligence.”

“Built for thinkers. Designed for creators.”

Clarity

Luchii makes complexity understandable.

Harmony

Luchii unifies information across domains.

Velocity

Luchii accelerates creation.

Frontier

Luchii pushes boundaries.

Week 1Teasers

Week 2Reveal

Week 3Deep-dive content

Week 4Developer onboarding

Developer API

One API. Drop-in ready.

One endpoint, four models. Bearer auth, familiar chat schema, and streaming responses. Public tier ships 60 req/min; enterprise scales to 600.

POST /v1/chat · /embeddings · /moderation
Models: 200m · 1b · 7b · 70b
Frasberg Public License (FPL)

Request access

POST https://api.frasberg.com/v1/chat
Authorization: Bearer <API_KEY>

{
  "model": "luchii-70b",
  "messages": [
    { "role": "user", "content": "Hello Luchii" }
  ],
  "max_tokens": 1024,
  "temperature": 0.7
}