aicanrun

AI Models for MacBook Neo A18 Pro — What runs on 8GB

14 great · 17 slow · 35 won't fit
Chip
Apple A18 Pro
Memory bandwidth
60 GB/s
Unified memory
8 GB
Usable for models
~5 GB
Cooling
Fanless
Year
2026

Specs checked against manufacturer and public documentation on . Apple does not publish a MacBook Neo memory-bandwidth figure; the speed range uses the A18 Pro chip profile and should be treated as lower-confidence.. Results below are estimates, not measurements.

What runs on the MacBook Neo A18 Pro

All 66 models at the best tracked quant for this 8GB configuration. Select any row for the full report and step-by-step setup guide.

ModelParamsQuantNeedsSpeedVerdict
Ternary Bonsai 4B4BQ2_02.2GB~22.9–31.6 tokens/s✓ Runs great
Llama 3.2 3B3.2BQ4_K_M3.1GB~12.6–17.4 tokens/s✓ Runs great
SmolLM3 3B3.1BQ4_K_M3GB~13.3–18.3 tokens/s✓ Runs great
Ministral 3 3B3BQ4_K_M3.2GB~12–16.6 tokens/s✓ Runs great
G9v3 3B3BQ4_K_M3GB~13.3–18.3 tokens/s✓ Runs great
Qwen 3.5 2B2BQ4_K_M2.3GB~19.4–26.8 tokens/s✓ Runs great
DeepSeek R1 Distill 1.5B1.8BQ4_K_M2.1GB~22.9–31.6 tokens/s✓ Runs great
Qwen3 1.7B1.7BQ8_02.8GB~14–19.3 tokens/s✓ Runs great
SmolLM2 1.7B1.7BQ4_K_M2.1GB~22.9–31.6 tokens/s✓ Runs great
Ternary Bonsai 1.7B1.7BQ2_01.4GB~50.4–69.6 tokens/s✓ Runs great
Llama 3.2 1B1.2BQ4_K_M1.7GB~31.5–43.5 tokens/s✓ Runs great
Gemma 3 1B1BQ4_K_M1.7GB~31.5–43.5 tokens/s✓ Runs great
OvisOCR2 0.8B0.8BQ4_K_M1.5GB~42–58 tokens/s✓ Runs great
Qwen3 0.6B0.6BQ8_01.5GB~42–58 tokens/s✓ Runs great
Bonsai 27B (1-bit)27BQ1_05GB~6.6–9.2 tokens/s! Runs, barely
Gemma 3 12B12.2BIQ1_Ssmaller version4.9GB~8.1–11.2 tokens/s! Runs, barely
Tini Cybersec 8B-A1B8.5BQ2_Ksmaller version4.7GB~69.1–95.4 tokens/s! Runs, barely
Llama 3.1 8B8BQ2_Ksmaller version4.7GB~7.9–10.9 tokens/s! Runs, barely
Ministral 8B8BQ2_Ksmaller version4.7GB~7.9–10.9 tokens/s! Runs, barely
Ternary Bonsai 8B8BQ2_03.7GB~11.5–15.8 tokens/s! Runs, barely
DeepSeek R1 Distill 7B7.6BQ2_Ksmaller version4.5GB~8.4–11.6 tokens/s! Runs, barely
Mistral 7B v0.37.2BQ3_K_Msmaller version5GB~7.2–9.9 tokens/s! Runs, barely
AREX Turbo 4B4.5BQ4_K_M4.2GB~8.7–12 tokens/s! Runs, barely
Fara 1.5 4B4.5BQ4_K_M4.2GB~8.7–12 tokens/s! Runs, barely
Gemma 3 4B4.3BQ4_K_M3.7GB~10.1–13.9 tokens/s! Runs, barely
Qwen3 4B4BQ4_K_M3.7GB~10.1–13.9 tokens/s! Runs, barely
Qwen 3.5 4B4BQ4_K_M3.9GB~9.3–12.9 tokens/s! Runs, barely
Nemotron 3 Nano 4B4BQ4_K_M4GB~9–12.4 tokens/s! Runs, barely
Agents-A1 4B4BQ4_K_M3.9GB~9.3–12.9 tokens/s! Runs, barely
Phi-4 Mini 3.8B3.8BQ4_K_M3.7GB~10.1–13.9 tokens/s! Runs, barely
Gemma 4 E2B2BQ4_K_M4.2GB~8.1–11.2 tokens/s! Runs, barely
Inkling952.4BQ8_0967.1GB—✕ Won't fit
Ornith 1.0 397B397BQ4_K_M285.8GB—✕ Won't fit
Hunyuan 3 (Hy3)298.8BQ4_K_M213GB—✕ Won't fit
Laguna S 2.1118BQ4_K_M80.7GB—✕ Won't fit
Llama 3.3 70B70BQ4_K_M50.3GB—✕ Won't fit
Qwen 3.5 35B-A3B35BQ4_K_M26.4GB—✕ Won't fit
Qwen 3.6 35B-A3B35BQ4_K_M24.1GB—✕ Won't fit
Ornith 1.0 35B-A3B35BQ4_K_M25.5GB—✕ Won't fit
KAT-Coder V2.5 Dev35BQ4_K_M25.7GB—✕ Won't fit
Laguna XS 2.133BQ4_K_M24.4GB—✕ Won't fit
Qwen3 32B32.8BQ4_K_M23.9GB—✕ Won't fit
Gemma 4 31B31BQ4_K_M22.2GB—✕ Won't fit
Qwen3 30B A3B30.5BQ4_K_M22.5GB—✕ Won't fit
Nemotron 3 Nano 30B-A3B30BQ4_K_M28.7GB—✕ Won't fit
Salience 1.5 Flash30BQ4_K_M22.5GB—✕ Won't fit
Qwen 3.5 27B27BQ4_K_M20.2GB—✕ Won't fit
Qwen 3.6 27B27BQ4_K_M18.7GB—✕ Won't fit
Ternary Bonsai 27B27BQ2_08.6GB—✕ Won't fit
Fara 1.5 27B27BQ4_K_M21.1GB—✕ Won't fit
Gemma 4 26B-A4B26BQ4_017.7GB—✕ Won't fit
GPT-OSS 20B21BMXFP415GB—✕ Won't fit
Qwen3 14B14.8BQ4_K_M11.3GB—✕ Won't fit
Phi-4 14B14.7BQ4_K_M11.4GB—✕ Won't fit
Ministral 3 14B14BQ4_K_M10.4GB—✕ Won't fit
Gemma 4 12B12BQ4_09GB—✕ Won't fit
Grug 12B12BQ4_K_M9.7GB—✕ Won't fit
grug 9B (ProCreations)9.4BQ4_K_M7.7GB—✕ Won't fit
Fara 1.5 9B9.4BQ4_K_M7.7GB—✕ Won't fit
Qwen 3.5 9B9BQ4_K_M7.4GB—✕ Won't fit
Ornith 1.0 9B9BQ4_K_M7.3GB—✕ Won't fit
Qwythos 9B v29BQ4_K_M7.4GB—✕ Won't fit
Qwen3 8B8.2BQ4_K_M6.6GB—✕ Won't fit
Ministral 3 8B8BQ4_K_M6.8GB—✕ Won't fit
LFM2.5 8B-A1B8BQ4_K_M6.8GB—✕ Won't fit
Gemma 4 E4B4BQ4_K_M6.3GB—✕ Won't fit

~ = bandwidth-based estimate · open a row to see the recommended app and exact steps

FAQ

What is the biggest local AI model the MacBook Neo A18 Pro · 8GB can run?

Bonsai 27B (1-bit) is the largest model in our 66-model comparison that fits at Q1_0. It needs about 5GB inside our 5GB working-memory budget, with an estimated 6.6–9.2 tokens/s decode range.

How much of the 8GB unified memory is available to a local LLM?

We use a conservative 5GB working budget, leaving room for macOS, the inference app, and normal background activity. Memory pressure, context length, and other open apps can change the real limit.

Can the MacBook Neo A18 Pro · 8GB run Llama 3.1 8B?

Yes. At Q2_K, our estimate uses 4.7GB and lands around 7.9–10.9 tokens/s.

Can the MacBook Neo A18 Pro · 8GB run Qwen 3.6 27B?

Not at the recommended Q4_K_M quant. It needs about 18.7GB in this 4K-context estimate.

Which app should I use for local AI on the MacBook Neo A18 Pro?

For an exact GGUF repository and quant from this site, start with LM Studio's graphical Discover, download, load, and chat flow. Jan is the open-source GGUF alternative; Ollama is strongest when the exact model already has a trustworthy catalog package or you want coding/API integrations; Msty is useful for mixed GGUF, MLX, and document workflows. For the curated Bonsai 27B build, try Locally AI and first confirm that its catalog shows the model on this Mac.

Can I upgrade the MacBook Neo A18 Pro to more unified memory later?

No. Apple-silicon unified memory is integrated into the chip package and must be chosen at purchase. Storage upgrades or external SSDs do not increase the memory an LLM can use.

Are these MacBook Neo A18 Pro local AI speeds measured?

No. Every speed on this page is a formula range based on Apple’s published memory bandwidth, model size, active parameters, and a cooling-aware efficiency range. App, backend, context length, and thermals can change real results.