This is wild. Google just dropped Gemma 4. Apache 2.0, open weights, frontier models that run on phones, laptops, and desktops👇

the "runs on phones and laptops" part is the real story. every previous open weights release was open in theory but required serious infrastructure. Gemma 4 changes what's actually deployable on-device, which is a different market entirely

🤔

Phones, laptops, desktops. Open weights. Apache 2.0. Fine-tunable. This is Google's real play - nt competing with ChatGPT for subscriptions, but making the model layer free and everywhere so every product built on top drives traffic back to Google's ecosystem.

👀👀

👀

Google here to take over (they already have)

They arent stopping

It makes me really happy ngl, my fav tool is notebookLM It is so underused and important

Come on, that’s not a frontier model 😅😅😅

The robotics angle is massive. Gemma 4 on edge = autonomous robots without cloud latency. China's Unitree already runs UnifoLM on-device. Japan's AIST has $180M in edge dexterity R&D. Lightweight open frontier models are the missing piece for Asia's humanoid deployment wave.

Gemma 4 on a local laptop is the end of the API bottleneck. There is nothing more soul-crushing than building a 3-minute video pipeline only to have it crawl because of API latency, cloud quotas, or a random 'Rate Limit' error right before a deadline. It kills the creative flow when you're waiting 60 seconds for a model to "reason" through a script change... Being able to run the full reasoning loop locally—zero lag, zero queues—is the ultimate freedom for solo-builders. We're moving from "rent-a-brain" to owning the entire studio on our own hardware.

frontier models on a phone. RIP to the battery life, but absolutely worth it.

Whats the benefit for normal humans?

This is the kind of drop that makes us stop scrolling immediately

Trying Gemma on my Mac M4 pro, I just found it a bit slower for the answers.

Putting frontier models in everyone’s hands changes the accessibility game completely. Open weights and device-level deployment mean experimentation and innovation can happen everywhere, not just in labs.

interesting tech, but let’s see real-world adoption first

It's only the beginning.

Apache 2.0 is simply fantastic. Here is my complete breakdown:

the cost curve for this is wild. had clients 6 months ago where the only viable option was API calls to frontier models. now some of those same tasks run locally for free. the procurement conversation with enterprises is completely different now.

26B MoE for low latency on device is the key spec. Apache 2.0 + fine-tunable means this runs in air-gapped enterprise deployments on day one.

Just wait until it’s the size of a microchip implanted in the brain ✨🧠✨

@grok why is this drop important?

It is trained to be the best open AI model 🔥

Apache 2.0 on phone hardware is a game changer tbh.

Apache 2.0 is the real headline here Google actually letting us build without the usual strings attached

If performance holds up, this could seriously push on-device AI forward.

Yes it is the way

On-device AI is going to change everything. Faster, cheaper, more private.