Skip to main content
Back

Google Unveils Gemma 3n Model That Runs On Your Phone

TRY NOW

May 22, 2025 4 min read

Hey,

Welcome to AlphaSignal – the most read newsletter by AI developers. 


We bring you the top 1% of news, papers, models, and repos, all summarized to keep you updated on the latest in AI.

IN TODAY'S SIGNAL

Read time: 4 min 15 sec

🎖️ Top News

📌 Speechmatics 

⚡️ Trending Signals

  💻 Top Lectures

🧠 Deep Dive

If you're enjoying AlphaSignal please forward this email to a colleague. 

It helps us keep this content free.

TOP NEWS

Edge AI

Google launches mobile-first Gemma 3n to process multimodal input without any internet connection

⇧ 11,389 Likes

Google has launched Gemma 3n, a compact AI model designed to run on your phone or laptop. It uses a new architecture optimized for efficiency, multimodal input, and the ability to operate offline. The model runs on devices with 2 GB of RAM while delivering 1.5x faster response than Gemma 3 4B.


Key Features

Gemma 3n introduces several technical innovations for mobile-first AI deployment.

  • The Mix’n’match architecture lets you switch between small and large models for cost and performance efficiency

  • Real-time multimodal processing that understands a mix of audio, video, and text inputs

  • With 446 tokens/sec processing on Samsung Galaxy S25 Ultra, you can build a fast, offline chatbot

Performance and Benchmarks
Gemma 3n ranks high on public model comparison platforms.

  • Ranks 2nd on Chatbot Arena with a score of 1283, just below Claude 3.7 Sonnet.

  • It has strong reasoning: Scores 64.9% on MMLU.

  • Good at languages: Scores 50.1% on WMT24++ machine translation benchmark, with improved understanding of Japanese, German, and Korean.

  • Fast on phones: Runs at 446 tokens/sec on GPU with under 5.5 GB memory use.

Access and Availability

You can use pre-trained weights and tokenizer directly or through preview tools.

  • Fine-tune using Hugging Face Transformers and QLoRA

  • Deploy via Google AI Studio (cloud) or AI Edge (on-device)

Try now

Building AI voice agents that never fumble the details

Just one missed detail—a name, number, or term—can derail a conversation.

Speechmatics offers:

  • Industry-leading and accurate speech-to-text solutions for enterprises

  • Flow Voice Agent API for reliable automation

The advanced voice AI stack precisely captures every critical detail in fast, multi-speaker, noisy environments in over 55 languages.

It’s built to handle complexity at scale, with:

  • On-prem, cloud, or on-device deployment

  • Custom dictionaries and intelligent formatting

  • Maintains low-latency, even during peak usage

Trusted by developers and global platforms.

Try for free

partner with us

TRENDING SIGNALS

AI Product Design

OpenAI acquires Jony Ive’s startup for $6.5 billion to create new AI-focused hardware, but it won’t possibly be a wearable

⇧ 28.492 Likes

Diffusion Model

Google’s new diffusion model generates text or code by gradually turning random noise into coherent output, like image diffusion models

⇧ 15,493 Likes

Coding Assistance

Mistral AI releases Devtral, its coding agent outperforming all open-source models such as DeepSeek R1 and Qwen 3 on SWE-Bench Verified

⇧ 4,205 Likes

AI Tool

xAI rolls out free beta of Live Search API for real-time internet and trending X content queries

⇧ 3,494 Like

APIs

OpenAI updates Responses API to create agents with image generation, code interpreter, and remote MCP server tool

⇧ 2,847 Like

TOP LECTURES

AI

Demis Hassabis talks about the frontiers of AI and the definition of AGI

⇧ 1,395 Likes

Nobel prize winner Demis Hassabis talks in this Google I/O session about frontier AI challenges and advances. He thinks AGI is a theoretical benchmark set by what a human brain’s architecture can do. Plus, you can learn about test-time compute, algorithmic scaling, Deep Think reasoning, and AGI definition boundaries.

AI Tools

A tutorial to build practical applications with the Gemini API 

⇧ 483 Likes

This session shows how to use the Gemini API for image parsing, live voice interfaces, calling APIs and agents through browsers, and document-based Q&A systems.

AI

Get highlights from Satya Nadella’s keynote at Microsoft Build 2025 within 15 minutes

⇧ 258 Likes

At the event, Nadella talked about open-source AI coding agents in VS Code, model fine-tuning with Copilot Studio and Foundry, and multi-agent workflows.

DEEP DIVE

AI Model

Insights into DeepSeek-V3: Scaling Challenges and Reflections on Hardware for AI Architectures

⇧ 4,538 Likes

This paper documents DeepSeek’s full-stack approach to training large language models, using 2,048 NVIDIA H800 GPUs.


Deepseek describes how it  FP8 precision, custom formats like LogFMT, and DeepEP to speed up training. It also uses fast network setups (RoCE, multi-rail fat tree) and a scalable file system (3FS) to move and store data efficiently.


The paper possibly gives you insight into building models when you have limited compute.

Read now

At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We’re passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.


Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We’re based in the United States. By using our site, you agree to be governed by U.S. laws.

Looking to promote your company, product, service, or event to 200,000+ AI developers? Let's work together.

Work with us