SearchSign inRequest Access
AltheonRequest access
ALTHEON

Sparse expert computation.

F2 uses sparse mixture-of-experts routing to expand model capacity without activating the entire parameter set for every token. Shared expertise remains available while routed experts specialize.

Efficiency is not a marketing claim. Every quantized or sparse configuration must survive the model's own promotion gates before it earns a capability label.

SynthesisSparse expert computation
Top-k routingShared expert
Load balanceExpert parallelism
01Top-k routing02Shared expert03Load balance04Expert parallelism
SurfaceModel Architecture
StatusPublic record
EvidenceTop-k routing / Shared expert
Next actionRequest access
Top-k routingShared expertLoad balanceExpert parallelism
TECHNICAL OBJECT

Altheon stays the model and system object.

Architecture, model cards, release state, and product proof remain connected without turning Berni, STRATA, STRATOS, or Aletheion into alternate names for the model.

SynthesisSparse expert computation
Top-k routingShared expert
Load balanceExpert parallelism
01Top-k routing02Shared expert03Load balance04Expert parallelism
ARCHITECTURE RECORD

Architecture and release boundary

The Altheon record exposes model identity, evidence class, evaluation boundary, and access path.

Router
Top-k routing, shared expert, balancing loss, capacity control, and expert-collapse monitoring.
Promotion gate
Sparse configurations are promoted only when regression, calibration, and hard-fail rates remain inside frozen thresholds.
Version family
F2 and related architecture route
Card linkage
Model card, system card, papers, and benchmark routes are linked from the page family.
01

Why MoE

Noetic tasks span science, code, finance, formal reasoning, language, and operational control. Sparse experts offer a path to specialization without monolithic activation cost.

02

Routing discipline

The router carries balancing and regularization objectives. Expert collapse is a measurable failure mode, not an implementation detail to ignore.

03

Scale profile

The Sovereign and Frontier configurations describe the released architecture families. Parameter counts, active compute, and evaluation evidence are kept explicit so capability claims remain tied to measured behavior.

ARCHITECTURE

Model identity, evidence, and release boundary.

Altheon records what belongs to the model, what belongs to surrounding systems, and which artifacts support each claim.

Current artifact

Router

Top-k routing, shared expert, balancing loss, capacity control, and expert-collapse monitoring.

Evidence to inspect

Promotion gate

Sparse configurations are promoted only when regression, calibration, and hard-fail rates remain inside frozen thresholds.

Operating boundary

What changes state

The record separates released claim, evidence class, boundary condition, and access path.

Access

Contact us for access

Use the access route for model review, benchmark packet, system walkthrough, or institutional collaboration.

EVIDENCE AND LIMITS

Public detail stays tied to source state.

Inspect evidence
Status
Public record
Primary artifact
Sparse expert computation.
Evidence class
Router
Source
Altheon
Date
2026-09-22
Related paper
Aletheon F2 White Paper / Yellow Paper
Benchmark
AMERICA-SIM plus related suites
Known limitation
Public pages exclude protected corpora, hidden cases, private scoring weights, partner data, and credentials.
Required next proof
Model card, system card, papers, and benchmark routes are linked from the page family.
ACCESS

Need access to this work?

For model architecture material, request access to the relevant card, benchmark packet, workflow demo, or review session.

Request accessCall the lab
RELATED
AltheonA frontier model built around noetic state, not answer production.Altheon F2The native F2 model family.Altheon ArchitectureDual Cortex cognition.Research ThesisNoetic intelligence.Decision Noetic IntelligenceThe released decision architecture of the left cortex.ArchitectureInter-Core Matrix.