A startup just put an entire fruit fly brain into a virtual body

While the rest of the AI industry is busy scaling transformer models to mimic human reasoning, one startup is trying a completely different playbook: rebuilding biological brains inside computers, neuron by neuron.

Eon Systems announced that it has connected a complete fruit fly brain emulation to a virtual body, producing multiple behaviors for the first time. The digital nervous system spans more than 125,000 neurons and over 50 million synapses.

According to co-founder Alex Wissner-Gross, the company mapped the fruit fly's neural wiring directly from electron microscopy data and plugged it into a simulated fly body running inside MuJoCo, the widely used physics engine.

Emulating biology instead of training networks

Unlike traditional AI labs, Eon is not training neural networks through reinforcement learning or statistical pattern matching. The team is trying to digitally copy real biological brains.

Previous whole-brain simulation attempts were far more modest. The well-known OpenWorm project tackled a nematode with just 302 neurons. Other virtual creature demos often relied on reinforcement learning to puppet a body, rather than simulating genuine biological brain dynamics.

Eon bypasses artificial training altogether by directly modeling the real biological circuit. And the fruit fly is just step one. The startup plans to emulate a mouse brain with 70 million neurons within two years, with the ultimate goal of simulating a human brain.

The brain code is out, but the body link is missing

There is an important catch to Eon's announcement.

While Eon uploaded code for its brain model to GitHub, that model is largely based on a 2024 Nature paper by Philip Shiu and colleagues. The startup has not yet published the code connecting that neural simulation to the virtual body in MuJoCo, which is precisely the breakthrough capability being claimed.

Until that bridge code is shared and peer-reviewed, researchers outside the company cannot verify how the virtual fly behaves or how reliably the neural signals translate into physical locomotion.

Why it matters

For researchers and roboticists, this announcement signals a radically different avenue toward embodied intelligence. Today's robotics industry pours billions into training end-to-end vision-language-action models, yet nature already solved sensorimotor coordination in tiny creatures with remarkable power efficiency. If digital connectomics can reproduce functional behavior directly from electron microscopy scans, robotics may eventually borrow blueprints directly from evolution.

We think Eon's conceptual positioning is fascinating. Instead of competing in the brutally expensive LLM compute race, it is hitching its wagon to the rapid advances in neuroimaging and physical simulation.

However, the technical leap from 125,000 neurons in an insect to 70 million in a rodent—let alone 86 billion in a human—is mind-bogglingly steep. Synaptic plasticity, neuromodulation, and precise biophysical modeling quickly become computational nightmares. Without the critical embodiment code publicly available, we remain cautious about whether Eon has truly solved neural control in a virtual body or simply staged an impressive conceptual milestone.