ZEBRAFISH

the recording

a real brain, watched in real time

google research spent years building the tools to read a brain. in 2025, they pointed a light sheet microscope at a larval zebrafish and captured everything.

70,634 neurons. every one tracked. nearly two hours of continuous thought, recorded through a transparent skull. this is not a connectome. it is a recording. the zapbench dataset does not map where the wires go. it shows what they do.

the same team is now mapping the structural connectome of this exact specimen. the wiring diagram and the recording of that wiring in action, from the same brain.

live

where the fish is right now

70,634 neurons running a leaky integrate-and-fire simulation driven by real calcium imaging data from the zapbench dataset. every number on this panel comes from the running model.

ZEBRAFISH BRAIN 70,634 NEURONS · ZAPBENCH DATA · LIF SIMULATION CALCIUM TECTUM CEREBELLUM STIMULUS 70,634 NEURONS 6,800 TIMEPOINTS ● LIVE

Neural activity

—

NEURONS ACTIVE of 70,634

mean ΔF/F

—

frame rate

~2 Hz

timepoint

—

stimulus

—

Brain region activity

Recording parameters

neurons

70,634

duration

~2 hrs

timepoints

~6,800

stimuli

14 types

calcium imaging via GCaMP. light sheet microscopy at cellular resolution. transparent skull, no surgery required.

Stimulus response

visual

—

motor

—

spontaneous

—

peak ΔF/F

—

google research

hhmi janelia

harvard

data: zapbench / ICLR paper / whole-brain calcium imaging, 70,634 neurons, apache 2.0

the signal

how a fish launched a token

the connectome is the map. the activity data is the mind. we fed both into a simulation and gave it one objective: propagate.

a vertebrate brain. not an insect, not a worm. a fish with a spine, optic tectum, cerebellum. it evaluated the network and found the shortest path to persistence.

it launched $ZFISH on robinhood chain through the pons v2 launchpad. one signed transaction. the first vertebrate to put something onchain.

the data

what the fish sees with

the zapbench recording captures activity across every major brain region in real time. these are the structures that drove the transaction.

Optic tectum

processes visual input. 14 stimulus types tested. the dominant region in a larval fish brain. receives direct retinal input and builds a spatial map of the world.

Cerebellum

motor coordination and learning. active during stimulus response. in larval zebrafish, already functional at 5 days post-fertilization.

Habenula

reward and aversion circuit. the fish's decision center. asymmetric between hemispheres — left and right encode different valences.

Hindbrain

motor output. drives swimming, turning, escape responses. the reticulospinal neurons here trigger the C-start escape reflex.

region annotations from the zapbench repository. calcium traces recorded at cellular resolution via light sheet microscopy.

the path

how it got here

not a roadmap of promises — a sequence, with everything before the amber line already done.

ZAPBench data loaded70,634 neurons, ~6,800 timepoints of calcium imaging across ~2 hours

done

Activity patterns extractedneural firing sequences isolated from calcium fluorescence traces (ΔF/F)

done

Visual system mappedoptic tectum responses to 14 stimulus types characterized

done

Simulation builtleaky integrate-and-fire model driven by real calcium activity data

done

Connected to chainrobinhood chain rpc, pons v2 launchpad integration

done

Token launched$ZFISH deployed via pons v2, one signed transaction from the simulation

done

Reward circuithabenula-raphe pathway, on-chain confirmation fed back as reward signal

building

on-chain

$ZFISH

launched by a vertebrate brain on robinhood chain. the first fish to put something onchain. read the receipt yourself.

Contract	0x96c393dee0e5dec75808675aebf00cf67f5d3778
Paired	GOOGL
Creator tax	—
Supply	—
Chain	Robinhood Chain · id 4663

open on the launchpad

the stack

what is running under the hood

70,634 neurons from the zapbench dataset. calcium imaging traces captured by light sheet microscopy at cellular resolution. every neuron in the larval brain, tracked for ~2 hours across 14 stimulus conditions.
leaky integrate-and-fire simulation. real activity patterns from the calcium traces drive the model. the simulation runs continuously, stepping through the recorded brain states.
habenula-raphe reward pathway. the zebrafish reward circuit maps on-chain confirmation to dopaminergic reinforcement. positive feedback from successful transactions strengthens the pathway.
pons v2 on robinhood chain. the simulation interfaces with the launchpad through standard RPC calls. one signed transaction deployed $ZFISH.
all source data is public. the zapbench dataset, the ICLR paper, and the google research blog post are linked throughout this site. verify everything.

the source

where the data comes from

every number on this site traces back to published, peer-reviewed research. nothing is invented.

data sources & references

dataset repository github.com/google-research/zapbench

ICLR publication ZAPBench: A Benchmark for Whole-Brain Activity Prediction

google research blog Mapping the Zebrafish Brain with ZAPBench

dataset documentation zapbench-release volume README