Products/AI/ML - Open Source Models/Google Gemma 4

Google Gemma 4

Byte for byte, the most capable open models

AI/ML - Open Source ModelsMountain View, United StatesFounded 1998Four model sizes: Effective 2B (E2B), Effective 4B (E4B), 26B Mixture of Experts (MoE), and 31B DenseAdvanced reasoning with multi-step planning and deep logicAgentic workflows with native function-calling, structured JSON output, and native system instructionsHigh-quality offline code generationMobile-first AI optimized for on-device useApache 2.0 open licenseRuns efficiently on Android devices, laptop GPUs, developer workstations, and acceleratorsFine-tunable for specific tasks
Visit Google Gemma 4 →
Google Gemma 4

Our Take

Google just dropped Gemma 4, and honestly? The numbers are absurd — the 31B dense model is out here ranking #3 in the world on Arena, outperforming models twenty times its size, and they're giving it away under Apache 2.0. The Gemmaverse thing is real, with over 100,000 variants spawned from 400 million downloads, which means developers have already voted with their compute. It's built from the same research as Gemini, runs on your laptop and Android phones, and actually ships usable reasoning without needing a data center — that’s the move.

Google's most intelligent open AI models to date, purpose-built for advanced reasoning and agentic workflows. Delivers unprecedented intelligence-per-parameter with breakthrough capabilities made widely accessible under an Apache 2.0 license.

Problem It Solves
Enables developers to achieve frontier-level AI capabilities with significantly less hardware overhead. Makes powerful open models accessible for on-device use, from billions of Android devices to developer workstations.
Target Customer
Developers, researchers, AI innovators, and organizations building AI-powered products
Use Cases
Building autonomous agents that interact with tools and APIs, Complex logic and reasoning tasks, Code generation, Fine-tuning for domain-specific applications (e.g., Bulgarian language model, cancer therapy research), On-device AI applications
Free Tier
true
Differentiator
Ranks #3 and #6 open model in the world on Arena AI text leaderboard. Outperforms models 20x its size. Built from same research as Gemini 3. Most intelligence-per-parameter in the industry.
Why Now
To build on community momentum - developers have downloaded Gemma over 400 million times creating 100,000+ variants in the Gemmaverse. Addresses demand for more capable open models that can run efficiently on various hardware.
Traction
User Count: 400000000 · Customers Mentioned: INSAIT (created BgGPT - Bulgarian-first language model), Yale University (Cell2Sentence-Scale project for cancer therapy research) · Notable Metrics: 31B model ranks #3 open model in world on Arena AI leaderboard; 26B model ranks #6; Over 100,000 variants in Gemmaverse

Key Facts

Category
AI/ML - Open Source Models
Location
Mountain View, United States
Founded
1998
Discovered via
product-hunt

The people behind Google Gemma 4

C

Clement Farabet

profile
O

Olivier Lacombe

profile

Links

WebsiteSource: product-hunt

Want products like this in your inbox every morning?

Five products. Every morning. Written by someone who actually cares whether they're good or not. Free forever, unsubscribe whenever.