The Bicameral Model: Bidirectional Hidden-State Coupling Between Parallel Language Models

2026-05-11

AI Summary

Traditional multi-model or tool-augmented systems rely on serialized textual communication, which is inefficient and constrains collaborative capabilities. This work proposes the first dual frozen language model architecture based on hidden-state coupling, establishing a bidirectional, continuous communication channel between intermediate layers via a trainable neural interface. A learnable inhibition gate is introduced to automatically discover efficient communication protocols without requiring predefined interaction formats. When integrated with tool backends—such as calculators, Z3 solvers, and Python sandboxes—the system boosts arithmetic task accuracy from 36% to 96%, achieves 1.7× the performance of baselines on logic puzzles, and enables the auxiliary model to generate problem-relevant code using only hidden-state signals.

Abstract

Existing multi-model and tool-augmented systems communicate by generating text, serializing every exchange through the output vocabulary. Can two pretrained language models instead coordinate through a continuous, concurrent channel? The Bicameral Model couples two frozen language models through a trainable neural interface on their intermediate hidden states. At every generation step, both models run in lockstep: a primary model drives the task while an auxiliary model operates tools, solves constraints, or executes code, with both conditioning on each other's activations through a translation network and a learned suppression gate ($\sim$1\% of combined parameters). The gate learns a selective communication protocol from task loss alone, without a prescribed format. We demonstrate the mechanism across three tool backends. On arithmetic, coupling two 0.5B models with a calculator raises accuracy from 36\% to 96\%. On logic grid puzzles, coupling two 0.6B models with a Z3 solver achieves $1.7\times$ the unaugmented baseline on ZebraLogic. On mathematical reasoning, coupling with a Python sandbox enables the auxiliary to generate problem-specific code from hidden-state signals alone, without ever seeing the problem text.

Problem

• language models
• tool-augmented systems
• hidden-state coupling
• model coordination
• continuous communication

Innovation

• Bicameral Model
• hidden-state coupling
• frozen language models
• selective communication
• tool-augmented reasoning