A new brain model called OmniMouse challenges conventional AI scaling assumptions by demonstrating that in neural activity modeling, data volume matters more than model size. The work, detailed in a preprint on arXiv, leverages a massive dataset of over 150 billion neural tokens recorded from the visual cortex of mice.
The Dataset and Model
The OmniMouse project uses recordings from 3.1 million neurons across 73 mice during 323 sessions. According to the researchers, the dataset includes neural responses to natural movies, images, parametric stimuli, and behavior, totaling more than 150 billion neural tokens—a token here being a recorded unit of neural activity.
OmniMouse is a multi-modal, multi-task model designed to operate in three regimes flexibly at test time:
- Neural prediction: forecasting future neural activity.
- Behavioral decoding: inferring behavior from neural signals.
- Neural forecasting: predicting upcoming neural dynamics.
The model can also combine these tasks. OmniMouse achieves state-of-the-art performance, outperforming specialized baselines across nearly all evaluation regimes, as reported by the authors.
| Dataset Feature | Value |
|---|---|
| Neurons recorded | 3.1 million |
| Mice used | 73 |
| Recording sessions | 323 |
| Total neural tokens | >150 billion |
Scaling Findings
The study’s central finding concerns how performance scales with data and model size. The researchers observed that performance scales reliably with more data, but gains from increasing model size saturate. This inverts the standard AI scaling story. In language and computer vision, massive datasets make parameter scaling the primary driver of progress. In brain modeling, however—even in the relatively simple mouse visual cortex—models remain data-limited despite vast recordings.
The authors note that this observation raises the possibility of phase transitions in neural modeling, where larger and richer datasets might unlock qualitatively new capabilities, paralleling emergent properties seen in large language models.
Implications for AI Research
While OmniMouse is focused on neuroscience, its scaling insights are relevant to enterprise AI leaders. The finding that model size saturates while data continues to improve performance suggests that for certain complex, high-variance domains—like brain activity or potentially supply chain sensor data—investing in larger datasets may yield greater returns than expanding model parameters. The code for OmniMouse is publicly available, enabling further research into scaling laws for biological and non-biological neural networks.