MODELS BY RAM
Best local AI models for 4 GB RAM
With 4 GB of RAM, you can run ultra-compact models ideal for basic chat, transcription, and voice tasks. These tiny models are optimized for minimal memory footprint while still delivering useful results.
48 models run · 7 too heavy · Low tier
FILTER BY TASK
Chat 15 models
Bonsai 8B (1-bit) 1.16 GB
PrismML's native 1-bit model — 8B params in 1.16 GB, needs forked llama.cpp
SmolLM2 135M 99 MB
Ultra-tiny language model — runs anywhere
SmolLM2 360M 267 MB
Ultra-small language model
Qwen 2.5 0.5B 477 MB
Lightweight chat — surprisingly capable
Gemma 3 1B 786 MB
Google's compact model — great reasoning for its size
SmolLM2 1.7B 1.0 GB
Largest SmolLM2 — punches above its weight
Qwen 2.5 7B 4.4 GB
Top-tier 7B chat — rivals much larger models
Llama 3.1 8B 4.9 GB
Meta's workhorse 8B — excellent all-around
Llama 3.2 3B 2.0 GB
Meta's 3B — best small model for many tasks
Gemma 3 4B 2.8 GB
Google's 4B — strong reasoning and instruction following
Llama 3.2 1B 791 MB
Meta's solid all-rounder
StableLM 2 1.6B 1.0 GB
Stability AI's efficient small model
TinyLlama 1.1B 780 MB
Popular tiny model trained on 3T tokens
Qwen 2.5 3B 2.0 GB
Sweet-spot model — great quality-to-size ratio
LFM2.5 1.2B 731 MB
Liquid AI's hybrid model — blazing fast CPU inference
Reasoning 6 models
DeepSeek R1 Distill 1.5B 1.1 GB
Distilled reasoning — chain-of-thought in a tiny package
Mistral 7B 4.3 GB
High-quality reasoning and analysis
DeepSeek R1 Distill 7B 4.7 GB
Distilled from DeepSeek R1 — strong step-by-step reasoning
DeepSeek R1 Distill 8B 4.9 GB
Llama-based R1 distill — best open reasoning at 8B
Qwen3 8B 5.2 GB
Alibaba's latest — thinking mode with strong reasoning
Phi-4 Mini 8.7 GB
Microsoft's reasoning model — exceptional for its size
Code 6 models
Qwen 2.5 Coder 0.5B 477 MB
Tiny code completion model — autocomplete on any device
Qwen 2.5 Coder 1.5B 1.1 GB
Best small code model — great for autocomplete
DeepSeek Coder 6.7B 3.8 GB
DeepSeek's code model — strong at generation and debugging
Qwen 2.5 Coder 7B 4.4 GB
Best open 7B code model — rivals GPT-4 on coding benchmarks
Qwen 2.5 Coder 3B 2.0 GB
Strong code generation and editing at 3B
StarCoder2 3B 1.9 GB
BigCode's multilingual code model — 600+ languages
Vision 4 models
SmolVLM 500M 490 MB
Tiny vision-language model — describe images on any device
LLaVA 1.6 7B 4.5 GB
Leading open vision model — image understanding and reasoning
Moondream 2B 1.3 GB
Small but capable vision model — image Q&A and captioning
Qwen 2.5 VL 7B 4.6 GB
State-of-the-art vision-language — image, video, document understanding
Embedding 4 models
Nomic Embed Text 137 MB
Open-source embedding with 8K context — long document search
GTE Large 335 MB
High-quality embeddings — top of MTEB benchmark at its size
all-MiniLM-L6-v2 23 MB
Fast sentence embeddings — ideal for semantic search
BGE Small 34 MB
Compact BAAI embedding — great for RAG pipelines
Image 3 models
Voice 4 models
Transcription 6 models
Wav2Vec2 Base 231 MB
High-accuracy English ASR
Whisper Small 488 MB
Good accuracy-speed balance — 99 languages
Whisper Tiny 89 MB
Real-time speech recognition — runs on anything
Whisper Medium 1.5 GB
Strong multilingual transcription
Distil-Whisper Large V3 1.5 GB
6x faster than Whisper Large — nearly same accuracy
Whisper Large V3 3.1 GB
Best open transcription — near-human accuracy
Too heavy 7 models
FLUX.1 Schnell 12.1 GB
Black Forest Labs' fast model — stunning quality in 4 steps
Gemma 3 12B 7.3 GB
Google's largest open model — near-frontier quality
Gemma 3n E2B 3.0 GB
Google's on-device multimodal — 2B effective params with vision and audio
Gemma 3n E4B 4.5 GB
Google's most capable on-device model — 4B effective, multimodal
Llama 4 Scout 63 GB
Meta's MoE model — 109B total params, 17B active per token
Mistral Nemo 12B 7.1 GB
Mistral's 12B — Tekken tokenizer, 128K context
Phi-4 Medium 8.0 GB
Microsoft's 14B reasoning model — frontier-class performance
BROWSE OTHER TIERS