F2 is the first native Altheon model line. Its current implementation is a causal sparse-MoE sequence model with hybrid attention, model-internal Dual Cortex supervision, control states, and native post-training paths.
F2 is the bridge from the sequence-model substrate we can train today to the noetic architecture Mourad developed. It does not pretend that bridge has already disappeared.
Architecture, model cards, release state, and product proof remain connected without turning Berni, STRATA, STRATOS, or Aletheion into alternate names for the model.
ARCHITECTURE RECORD
Architecture and release boundary
The Altheon record exposes model identity, evidence class, evaluation boundary, and access path.
Architecture
Sparse MoE sequence model with grouped-query attention, rotary encoding, Q/K normalization, SwiGLU, shared expert, and routed experts.
Release identity
F2 is presented as the native model line, with its card, training path, and evaluation boundaries made visible.
Version family
F2 and related architecture route
Card linkage
Model card, system card, papers, and benchmark routes are linked from the page family.
01
Implemented core
F2 includes grouped-query attention, rotary position encoding, Q/K normalization, SwiGLU, sparse routed experts, a shared expert, multi-token prediction, and learned Dual Cortex fusion.
02
Control Cortex
The control branch is trained against SUPPORTED, UNSUPPORTED, CONTRADICTED, and UNKNOWN states. The objective is to give the model an internal representation of epistemic condition, not merely a post-hoc filter.
03
Research direction
The release program moves more of the authoritative cognition into structured noetic state and tests whether that shift improves discovery yield, belief repair, invariance, and decision discipline under matched resources.
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
Architecture
Sparse MoE sequence model with grouped-query attention, rotary encoding, Q/K normalization, SwiGLU, shared expert, and routed experts.
Evidence to inspect
Release identity
F2 is presented as the native model line, with its card, training path, and evaluation boundaries made visible.
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.
Public pages exclude protected corpora, hidden cases, private scoring weights, partner data, and credentials.
Benchmark claims stay tied to named workload, version, comparator, and recorded conditions.
Release authority, tool authority, and answer generation are separate control surfaces.
Inspect evidence
Status
Public record
Primary artifact
The native F2 model family.
Evidence class
Architecture
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 altheon f2 material, request access to the relevant card, benchmark packet, workflow demo, or review session.
