Introducing the newest Coral board, for efficient, on-device AI! Check out the demos in the video: - On-board speech translation - Natural language controlling hardware - Vision & sound generating music

Coralboard arrives this summer! ☀️ Learn more about the edge AI platform at or get ready to build by checking out the Coralboard setup guide at

I’m going to add this to a drone with a nerf gun and battle my friends.

But we already have Gemma 4. People want power to run Gemma 4, not Gemma 3

It is $129.99 for those who want to know its price.

Our accelerator engine at @apexcompute is already running the Gemma 3 1B model at 10 tokens/s on an FPGA prototype while consuming under 5W. The ASIC version coming later this year will deliver even higher performance with even lower power consumption for edge AI applications.

really cool, but first french translation is messed up

$GOOG and $NVDA are all trying to eat Raspberry Pi’s market😂😂

the real test for on-device ai isn't the demo, it's getting it to run reliably on 100k mediatek phones with indian accents and low-light cameras. your jetson nano struggles with this in mumbai traffic

no thanks, i’m already running mine on an 8086

On-board speech + vision + hardware control with Gemma locally — this is the future of edge AI. For builders experimenting with on-device prompts, here's one I use for natural language hardware control workflows:

Proud to have won this 🏆 at Google I/O 🎉, after completing the adventure quest 🗺️😊

I look forward to being able to give these to my kids to see what they can come up with, because they don't care about what is possible or correct, only what in the end they can make happen.

This will be really good for building small devices with AI. What's the price? Would it be possible to upgrade the model in future?

@grok compare it to Raspberry Pi or Arduino

on-device AI: great until the model needs a patch and the board is in a remote sensor 400km away.

It is the era of on-device AI! But the security side of it is also important. More details can be found here:

@grok what can 1 TOP run locally without feeling like 2019?

I clearly need one :)

No, thank you. I run Gemma 4 31B locally on my GPU. I don't need this crap.

Why not start with the most advanced model that can do everything?

Can we get one that runs Gemma 4 31b?

@grok It seems Google forgot to add easily accessible information about pricing and how to get the device. Beginner level marketing error, can you help getting me that information?

Curious how far we are from local-first AI agents becoming the default architecture cloud for training / edge devices for execution

this is so cool! what's the price gonna be?

Bring it on already! We are strapping Raspberry PIs with AI hats and Nvidia jetson orin nano super running on our edge devices. I have been scouring your shop for ages. I do have the 1Gen USB device.

Is there a waitlist somewhere so I can be notified when units become available?

i'd try the natural language controlling hardware demo first, but i don't trust it to hold up once commands get sloppy, which is usually where things break.

Price not being mentioned is a bad sign. $120 USD? 🤔

@grok What are some of the most innovative use cases you foresee for this new Coral board? 🛠️🚀 #EdgeAI #GoogleGemma

how can i get my hands on one?? @googlegemma @googleaidevs

This is the interesting edge AI pattern: not replacing cloud models, but deciding locally when the cloud is even needed. The hard parts for products will be power budgets, sensor noise, tool permissions, and fallback UX when the tiny model is uncertain.

@osanseviero Hah. I was just making a food pantry home server thing with Gemma 4. Coral board would slot in perfectly.

this is what the people want! let me build an r2d2 that i can talk to! and it just beeps at me

put that into Reachy mini would be insane