ZEHEN Labs

The structure
underneath.

We find mathematical structure — phase transitions, coupling dynamics, free energy landscapes — hiding inside complex systems. Then we build tools on it. Physics, network theory, information theory, dynamical systems. Whatever the problem needs.

Nc ~ 3.5B
Median Critical Scale
102 models
63 Base + 39 Frontier
r = −0.989 (Pythia)
Pre-transition
NeurIPS 2026
2 Papers Submitted
Explore the Dashboard → Read the Papers
CAPE

Capability Coupling Analysis
of Phase Emergence

Pick two capabilities that should be independent. Measure their coupling as a model scales. Find where coupling changes sign. That sign-flip is a phase transition — and it tells you everything about what your model can and can’t do.

Below Nc
Alignment taxes capabilities. Nc varies by family and architecture — ~3.5B median for standard training, lower or absent for curated models.
At Nc = 3.5B
Critical point. Maximum susceptibility. Small interventions have maximum leverage.
Above Nc
Capabilities cooperate. Scale freely. The same pattern repeats at every cascade level.
Nc = 3.5B
Critical Scale
r = −0.989 (Pythia)
Pre-transition Coupling
0.513
Frontier Slope
5.6%
ODE Cross-Prediction MAE
63 + 39
Base + Frontier Models
16
Model Families
10
Frontier Labs
7
Falsifiable Predictions
Nc = 3.5B
Critical Scale
r = −0.989 (Pythia)
Pre-transition Coupling
0.513
Frontier Slope
5.6%
ODE Cross-Prediction MAE
63 + 39
Base + Frontier Models
16
Model Families
10
Frontier Labs
7
Falsifiable Predictions
The Discovery

Watch capabilities flip from fighting to cooperating

Each dot is a model. As scale increases (left to right), the coupling between reasoning and truthfulness crosses zero. Below: they fight. Above: they cooperate. Every family shows this.

The coupling sign flip (Paper 3A) transitions into the frontier landscape (Paper 3B). Same physics, different scale. Read the full story →

Dashboard

Try It

Enter your model’s benchmarks. Get its phase, coupling trajectory, and what to do next.

CAPE Dashboard
Phase classification, h-field diagnostic, ODE trajectory fitting, frontier coupling analysis, self-steering demo, and 7 falsifiable predictions. Works for any model from 70M to frontier scale.
63 + 39 models
12 tools
10 labs
Open Dashboard →
Products

Tools We’re Building

Basin Memory

PRE-RELEASE

Memory as an energy landscape. Retrieval is Boltzmann-weighted — temperature controls whether you explore (high T, creative) or exploit (low T, precise). Memories deepen with use. Offline consolidation merges, prunes, and strengthens — the same dynamics as biological sleep.

Not just for agents. The energy landscape applies to any system with persistent memory: conversational AI, knowledge bases, research tools, clinical note systems, education platforms. The physics is domain-agnostic.

100%
LoCoMo Judge
100%
LongMemEval
158ms
Latency
98.1
BasinBench (ours)
Beats Mem0 (49.7%), OpenClaw (72.5%), PropMem (82.3%) on LoCoMo. 9 physics signals. Hebbian deepening. Kramers escape rates for forgetting dynamics.
Request early access →
$ pip install basin-memory

>>> from basin_memory import Basin
>>> b = Basin(temperature=0.7)
>>> b.remember(doc, context)
>>> b.retrieve(query)
>>> b.sleep() # consolidate
EMNLP ARR · May 25 deadline
Paper: “Physics-Informed Agent Memory with Dynamic Free Energy Learning”

CAPE Dashboard

LIVE

Phase classification, h-field diagnostic, ODE trajectory fitting, self-steering demo. Enter benchmarks, get actionable guidance.

Open Dashboard →

cape-steer

OPEN SOURCE

Activation-level alignment correction for any open-weight model. Auto-detects architecture, finds the coupling bottleneck, steers at quarter-depth. Works on CPU.

GitHub →
Papers

Current Work

Two papers submitted to NeurIPS 2026. More in preparation across multiple domains.

3A
“Lying Is Just a Phase” — The Hidden Alignment Transition in Language Model Scaling
NeurIPS 2026
Submitted
3B
“The Growing Pains of Frontier Models” — Capability Coupling at Frontier Scale
NeurIPS 2026
Submitted
All Papers → Read the Blog Post →
About

ZEHEN Labs

We look for mathematical structure in complex systems — drawing from physics, dynamical systems, network theory, information theory, and whatever else the problem needs. When we find structure, we build tools on it.

Current focus: AI scaling laws. We discovered that the coupling between model capabilities undergoes a phase transition at a critical scale, and that transition is predictable, measurable, and actionable.

Founded by Adil Amin. Based in Milwaukee, WI.

ذہن
ze·hen  /ˈzɛ.hɛn/
mind · intellect · understanding
Urdu · Persian

Get in Touch

Interested in collaboration, consulting, preprints, or early access?

adil@zehenlabs.com GitHub