Small language models, held to the same standard. Efficiency isn't an excuse for opacity.
Models
13
CPU-Only Models
12
Min RAM Required
0.5GB
Frameworks
6
Why SLMs for Local Dev?
You don't need a data center to build AI-powered apps. These models run on a laptop, a Raspberry Pi, or even in a browser — no GPU required, no cloud costs, no API keys.
Get Started in Minutes
Install Ollama or LM Studio, pick a model below, and run the quick-start command. Your first local LLM in under 5 minutes.
Privacy by Default
All inference runs locally. Your data never leaves your machine — ideal for sensitive domains, offline environments, or just personal projects.
13 models — sorted by parameter count (smallest first)
Runs in a browser via WebAssembly. The go-to for truly embedded or browser-native AI experiences.
Half a billion parameters — the most resource-efficient model in the catalog. Surprisingly capable for simple tasks.
Google's latest ultra-compact Gemma 4 with multimodal vision support. Designed for edge and on-device deployment.
The smallest Gemma 3 — perfect for truly constrained environments. Runs on a Raspberry Pi 5.
Meta's official 1B model with an extraordinary 128K context window. Optimized for on-device and mobile deployment.
The classic entry-point SLM. Extremely fast on CPU, minimal RAM usage — great for learning and prototyping pipelines.
A solid 1.5B model with surprisingly strong multilingual capabilities and a 32K context window.
HuggingFace's best-in-class small model for its size. Trained on high-quality curated data for impressive coherence.
Excellent instruction-following at 3B. One of the best multilingual SLMs for local development.
Meta's 3B model balances capability and footprint well. Great starting point for RAG and agentic pipelines on local hardware.
Exceptionally capable small model from Microsoft. Punches well above its weight class on reasoning and coding tasks.
Microsoft's compact powerhouse with a massive 128K context window. Ideal for long-document tasks on consumer hardware.
Balanced multimodal SLM from the Gemma 4 family. Strong vision-language understanding in a 4B footprint.