AI PLATFORMS & MODELS
SMALL / EFFICIENTOPEN SOURCE SELF-HOSTABLE

Gemma 3

by Google DeepMind

Open models from 1B to 27B, single-GPU friendly.

84HYLAND SCORE
5.3k views this week
Try Gemma 3
SHARE Post on X LinkedIn

Overview

Gemma 3 is Google’s open model family spanning 1B, 4B, 12B, and 27B sizes with multimodal support and a 128K context. Designed to run on a single GPU or even a phone, with a permissive license.

Key features

  • Four sizes: 1B / 4B / 12B / 27B
  • Vision support (4B+)
  • 128K context window
  • Runs on a single GPU
  • 140+ languages

Tags

Self-HostEdgeMultimodalSingle GPU

Related in AI Platforms & Models

VERIFIED

Gemini 2.0 Flash

Google DeepMind
87HYLAND

Workhorse multimodal model at Flash speed.

Small / EfficientFreemium
6.8k/wk
View
VERIFIED

GPT-4o mini

OpenAI
85HYLAND

Fast, cheap, and surprisingly capable.

Small / EfficientFreemium
7.2k/wk
View
VERIFIED

Claude 3.5 Haiku

Anthropic
84HYLAND

Anthropic’s fastest model for scaled workloads.

Small / EfficientPaid
4.0k/wk
View