#1
LFM2.5-8B-A1B
Liquid AI hybrid model built for on-device assistants. 8.3B total / 1.5B active, 128K context, tool use, GGUF, ONNX, MLX, llama.cpp and LM Studio support. Open-weight under LFM 1.0.
chatcodereasoningspeedstandard
RAM tier guide
Best local LLMs for 8GB RAM
A static, Google-indexable guide to the best local AI models that fit in a 8GB RAM budget. Built from the LocalClaw model database and ranked by quality, reasoning, coding and speed.
Compatible models
78
Best pick
LFM2.5-8B-A1B
RAM tier
8GB
Hardware fit
entry-level laptops, MacBook Air 8GB, compact PCs and everyday local AI experiments
Quick answer
With 8GB RAM, prioritize models with minimum RAM at or below 8GB and avoid filling memory completely. For most users, start with LFM2.5-8B-A1B, then test a faster smaller model if latency matters.
Top models for 8GB RAM
#2
Granite 4.1 (8B)
IBM Granite 4.1 long-context instruct model. Apache 2.0, 131K context, tool calling, RAG, code tasks, multilingual dialog and business assistant workflows on normal 8-16 GB machines.
chatcodereasoningstandardgeneral
#3
DeepSeek R1 0528 Distill (8B)
Updated R1 reasoning distilled to Qwen3-8B. Improved chain-of-thought with fewer hallucinations vs original R1 distills. MIT licensed.
reasoningstandard
#4
Qwen 3.6 (6.7B)
Alibaba's hybrid-thinking micro-flagship. Toggles between instant answers and deep chain-of-thought reasoning on demand. 128K context, 29 languages, outperforms Qwen3-8B on reasoning benchmarks. Apache 2.0.
chatcodereasoningspeedgeneral
#5
Llama-3.1-Nemotron-Nano (4B)
⭐ Mac Mini M4 16GB top pick! NVIDIA fine-tune of Llama 3.1. Hybrid /think • /no_think mode — deep reasoning on demand, instant chat otherwise. ~80–120 tok/s on Apple Silicon Metal. 128K context. Apache 2.0.
chatlightspeedreasoning
#6
Qwen 3 (8B)
One of the best 8B models ever made. Thinking mode + lightning fast. The new king of 8B.
chatcodestandardgeneralreasoning
#7
Nemotron 3 Nano (4B)
⭐ Mac Mini M4 16GB top pick! NVIDIA's hybrid model — distilled from 9B, keeps 95% of its quality. Hybrid attention + SSM layers = ~80–120 tok/s on Apple Silicon. Blazing fast, minimal RAM. NVIDIA Open Model License.
chatlightspeedreasoning
#8
Qwen 3.5 (9B)
The best small Qwen 3.5 for everyday use. Strong reasoning, coding and chat at 9B scale with hybrid thinking mode and 256K context. Runs on 8-16 GB RAM. Great for Mac Mini M4 Pro. Apache 2.0.
chatcodereasoninggeneral
#9
Granite 3.3 (8B Instruct)
IBM enterprise-grade 8B. Trained for RAG, tool-use and structured output. Strong function calling and long-context performance (128K). Apache 2.0 with full data provenance.
chatcodestandardgeneralreasoning
#10
Gemma 4 E4B
Gemma 4 balanced edge model with strong multimodal quality and 256K context. Great for laptops and high-end mobile devices. Apache 2.0.
chatvisionstandardmultimodalreasoning
#11
DANTE-Mosaic-3.5B
OdaxAI compact dense model based on SmolLM3-3B and distilled from Kimi K2. Strong small-model benchmark profile: GSM8K 74.45, HellaSwag 76.73 and MBPP 42.6. Apache 2.0, BF16 weights, practical for local Transformers/vLLM use.
chatreasoningcodelightmultilingual
#12
Qwen 3 Coder (8B)
Qwen coding specialist with long context. Great for agentic coding tasks. 477K downloads.
codestandard
#13
Gemma 3n (8B)
Google on-device powerhouse with vision. Designed for phones/tablets/laptops but punches far above its weight. Per-layer memory management for constrained devices. Apache 2.0.
chatvisionstandardgeneral
#14
DeepSeek R1 Distill (8B)
DeepSeek's reasoning model distilled to 8B. Shows its thought process step-by-step. Mind-blowing for logic.
chatreasoningstandard
#15
Cogito (8B)
Hybrid reasoning model outperforming peers. Strong general + reasoning at 8B. 558K downloads.
chatreasoningstandardgeneral
#16
EXAONE Deep (7.8B)
LG AI Research reasoning model. Strong at math and coding reasoning. 200K downloads.
reasoningstandard
#17
Qwen 2.5 (7B)
Alibaba's 18T token trained model. Excellent multilingual and coding. 14.9M downloads. Wide community support.
chatcodestandardgeneral
#18
InternVL3 (8B)
Shanghai AI Lab multimodal model. Strong vision understanding for documents, charts, and photos. MIT licensed. Note: primarily PyTorch/safetensors — community GGUF may vary.
visionstandard
How to choose at 8GB
- Use Q4_K_M or Q5_K_M quantization for the best quality/speed balance.
- Leave memory headroom for macOS/Windows, browser tabs and LM Studio overhead.
- For coding, prioritize coding and reasoning scores. For chat, quality and speed matter more.
- If a model feels slow, drop one size tier before changing hardware.