MacBook Air M3 8GB has 8GB of unified memory and is a strong fit for portable local LLM experiments. These recommendations are generated from the current LocalClaw catalogue and filtered for realistic memory headroom.
Recommendations updated September 8, 2026
Start with MiniCPM5 2B on this Mac. A comfortable or good fit leaves useful memory for macOS and your local runtime. A tight fit can still work, but close other apps, reduce context length when needed, and prefer the listed quantization.
Official OpenBMB compact on-device LLM with Apache 2.0 licensing, 131K context, tool-calling and coding focus, plus official Q4_K_M GGUF, Ollama, llama.cpp, Docker and OpenClaw run paths.
IBM Granite 4.2 3B is the compact Apache 2.0 Granite reasoning model with 128K native context, thinking-mode chat, tool calling and official GGUF artifacts for laptop-class local inference.
Liquid AI edge vision-language model with LFM2.5-2.6B backbone, SigLIP2 NaFlex vision encoder, 32K context, LFM 1.0 open weights and official GGUF plus llama.cpp and MLX runtime paths for local image chat and OCR.
Liquid AI compact hybrid model with 128K context, LFM 1.0 open weights, official GGUF, ONNX and MLX artifacts, and practical llama.cpp / LM Studio paths for 8GB-class local machines.
DFM-Mimir is a Danish and English 1B-class HRM language model trained from scratch by Danish Foundation Models, with Apache 2.0 weights, permissible-data positioning and GGUF artifacts for lightweight local inference.
InternScience compact dense agent model with Apache 2.0 licensing, 262K context and official Q4_K_M GGUF artifacts for 8GB-class local assistants.
IBM Granite 4.1 compact long-context instruct model. Apache 2.0, 131K context, tool calling, RAG and code tasks, with an official Q4_K_M GGUF for practical 4-8 GB local machines.
NVIDIA fine-tune of Llama 3.1 with hybrid /think and /no_think modes, 128K context and Apache 2.0 licensing. A compact local option for reasoning and chat.
NVIDIA compact hybrid model distilled from a 9B teacher, with hybrid attention and SSM layers. A lightweight option for local chat and reasoning under the NVIDIA Open Model License.
The shared LocalClaw engine first rejects hosted-only, excluded and oversized records. It reserves system and 8k-context headroom, labels comfortable, good and tight fits, then ranks the remaining models by hardware fit, use case, catalogue capability ratings, runtime and freshness. Community stars are never included. This is practical guidance, not a standardized third-party benchmark.
This guide is about local AI fit, not live pricing. Prices and availability change. An Amazon link may be an affiliate link that supports LocalClaw at no extra cost.