- Horizon AI
- Posts
- How to Run Google's New Gemma Model Locally on Your Phone 📱
How to Run Google's New Gemma Model Locally on Your Phone 📱
Google Drops Gemma 4: The Most Intelligent Open Model Ever? 🤯
Gina Acosta
April 06, 2026
In partnership with
Welcome to another edition of Horizon AI,
Running AI models locally means no internet required and total privacy for your prompts, images, and sensitive data. In today's issue, we show you how to run Google's new open-source models directly on your phone in minutes, for free.
Let’s jump into it!
Read Time: 4.5 min
Here's what's new today in the Horizon AI
Google Releases Gemma 4, Now Fully Open Source
Anthropic Cuts Off Third-Party Tools Like OpenClaw for Claude Subscribers
AI Tutorial: How to run Google's Gemma 4 model locally on your phone
AI Tools to check out
AI Findings/Resources
The Latest in AI and Tech 💡
AI News
Google Releases Gemma 4, Now Fully Open Source
Google has introduced Gemma 4, a new family of open models with larger variants for powerful servers and smaller ones designed to run efficiently on phones, PCs, and edge devices.
Details:
Earlier versions of Gemma were released under a custom license that came with usage restrictions and terms Google could update at will, causing many devs to be apprehensive about building with them.
Now, Gemma 4 is being released under an Apache 2.0 license, which is much more permissive, with no overbearing terms of use or commercial restrictions.
Gemma 4 comes with improvements in reasoning, math, instruction-following, code generation, and visual input processing, all across more than 140 languages.
For edge devices including smartphones, Google offers the 2-billion (E2B) and 4-billion (E4B) "Effective" models, designed to use less memory and battery than Gemma 3, with Google touting "near-zero latency."
For more powerful machines, there's the 26-billion "Mixture of Experts" and 31-billion "Dense" models.
Context windows have also expanded, with edge models now supporting 128k tokens and the 26B and 31B models getting 256k.
Google claims that so far "developers have downloaded Gemma over 400 million times, building a vibrant Gemmaverse of more than 100,000 variants." Now that Gemma 4 is being released as pure open-source software, the company is hoping adoption rates will pick up even more.
TOGETHER WITH OUTSKILL
Build real AI automations this weekend. Live. For free.
Still figuring out how to actually upskill in AI this year? You're not alone. But 48% of workers already used AI tools in 2025, and the gap is only growing.
Outskill's free 2-day live mastermind walks you through building AI agents, automating your workflows, and turning AI skills into real income.
🗓️ 16 hours, Saturday and Sunday, 10 AM to 7 PM EST.
Show up, and you'll also unlock a Prompt Bible, an AI monetization roadmap, and a personalized toolkit builder. All free.
ANTHROPIC
Anthropic Cuts Off Third-Party Tools Like OpenClaw for Claude Subscribers
Anthropic is changing how third-party tools like OpenClaw work with Claude, making usage more expensive for subscribers.
Details:
Claude subscriptions no longer cover usage through third-party tools like OpenClaw, forcing users onto separate pay-as-you-go pricing.
Users can still access OpenClaw with Claude, but only by purchasing additional usage bundles or using the API, increasing overall costs.
Anthropic says the change is due to high demand and infrastructure limits, as third-party tools consume more resources than subscriptions were designed for.
The company is offering a one-time credit equal to a user’s monthly plan, along with discounted usage bundles as a transition.
Peter Steinberger, OpenClaw creator recently hired by rival OpenAI, said he and board member Dave Morin "tried to talk sense into Anthropic, the best we managed was delaying this for a week." The move appears aimed at steering users toward Anthropic's own tools, like Claude Cowork, instead.
AI Tutorial
How to run Google's Gemma 4 model locally on your phone
Download the Google AI Edge Gallery from the Google Play Store, App Store, or install the APK from the latest release on GitHub.
Open Google AI Edge Gallery, tap on AI Chat, and download Google's Gemma 4 model E2B or E4B. (Pick the variant that best fits your device. The heaviest model isn't always the best choice.)
You can also click on the “+” at the bottom to import your own models.
Once downloaded, you're ready to run everything locally on your phone. No internet connection needed.
Select AI Chat to start a conversation, or explore the other available features:
Agent Skills: transforms your LLM from a conversationalist into a proactive assistant.
Ask Image: use multimodal power to identify objects, solve visual puzzles, or get detailed descriptions using your camera or photo gallery.
Audio Scribe: transcribe and translate voice recordings into text in real time.
AI Tools to check out
🚀 Krev: AI creative agents for ecommerce brands. Generate branded ads, studio-quality product photos, and videos in minutes
🎙 Noiz: A next-generation voice platform that makes it easy for anyone to create natural, expressive, and studio-quality voices.
⭐ Influcio: AI marketing agent for result-driven influencer campaigns.
🦾 Panorama: It analyzes your workplace data to recommend hidden structures and AI workflows your team can run together.
🛠 Xcode: Everything you need to develop, test, and distribute apps across all Apple platforms.
TOGETHER WITH MINTLIFY
Are you tracking agent views on your docs?
AI agents already outnumber human visitors to your docs — now you can track them.
AI Findings/Resources
🤯 What teens are doing with those role-playing chatbots: Harassing bots with “funny violence,” confiding in them, and even dating.
📷 Reddit user asks ChatGPT to make a photo of a college party in 2004 taken on a flip phone
🍻 How an Irish genius drove down the price of Guinness using AI modeled after reality TV winner
🚀 OpenAI co-founder AndrejKarpathy shares 'LLM Knowledge Base' architecture that bypasses RAG with an evolving markdown library maintained by AI
The latest in AI and Tech
Netflix has introduced an AI model called VOID that can remove objects from video scenes and realistically rebuild them. The tool goes beyond basic editing by understanding interactions and generating natural-looking results after changes.
This new model allows users to have a face-to-face, real-time conversation with any AI agent by allowing them to join Google Meet calls.
The company is reshuffling its leadership team, with COO Brad Lightcap moving into a new role focused on "special projects" and strategic deals. Other executives are also shifting responsibilities, while some, like Fidji Simo, are stepping back temporarily for health reasons.
Microsoft has released three new AI models: MAI-Transcribe-1, which transcribes speech across 25 languages; MAI-Voice-1, an audio-generating model that produces 60 seconds of audio in one second and supports custom voice creation; and MAI-Image-2, a video-generating model. All three are available on Microsoft Foundry, with the transcription and voice models also accessible through MAI Playground.
That’s a wrap!
We'd love to hear your thoughts on today's email!Your feedback helps us improve our content |
Not subscribed yet? Sign up here and send it to a colleague or friend!
See you in our next edition!
Gina 👩🏻💻