Yes, a small one. The MacBook Neo has Apple's A18 Pro chip and 8 GB of unified memory, the only size Apple lists, and memory is what decides which local models fit. About 6 GB of it is available to a model, so Qwen 3.5 2B and Grux Local 2B rate a Great fit and Llama 3.2 3B, Qwen 3.5 4B and Grux Local 4B rate a Good fit; the rest of Grux's catalogue is rated in the table below. Start with ollama run qwen3.5:4b or ollama run qwen3.5:2b. Whether Ollama or Grux AI runs on this chip has not been tested here: Ollama's macOS page names Apple M series chips, and the A18 Pro is not one, and we have not run either on a MacBook Neo. This page quotes no speed, because nothing on it was measured on a MacBook Neo.
| Part | What Apple's tech specs say |
|---|---|
| Chip | Apple A18 Pro chip, with a 16-core Neural Engine |
| Memory | 8GB unified memory, the only memory size listed |
| Memory bandwidth | 60GB/s |
| Storage | 256GB or 512GB SSD |
Unified memory is shared by macOS, your apps and the model, which is why 8 GB is the number to plan around. Models are also large files: see where Ollama stores them and how to move or remove them before filling a 256GB drive.
Each rating uses the which LLM can my Mac run page's rule: the memory a model needs while it is running, which is usually more than its file, against about 75 percent of a Mac's unified memory, which is 6 GB of the 8 GB here. The table is generated from that rule, the same one the Mac picker uses.
| Model in Grux's catalogue | Memory it needs while running | Rating on an 8 GB Mac | Ollama needed |
|---|---|---|---|
| Grux Local 2B | 2.65 GB | Great fit | Ollama 0.40.8 or newer |
| Grux Local 4B | 4.44 GB | Good fit | Ollama 0.40.8 or newer |
| Qwen 3.5 2B | 2.6 GB | Great fit | Ollama 0.30.0 or newer |
| Qwen 3.5 4B | 4.4 GB | Good fit | Ollama 0.30.0 or newer |
| Llama 3.2 3B | 3 GB | Good fit | no minimum listed |
| Qwen 3 VL 4B | 6 GB | Too big | Ollama 0.30.0 or newer |
| Grux Local 9B | 7.25 GB | Too big | Ollama 0.40.8 or newer |
| Qwen 3.5 9B | 6.2 GB | Too big | Ollama 0.30.0 or newer |
| Gemma 4 12B | 9.4 GB | Too big | Ollama 0.30.9 or newer |
| Mistral Small 3.2 24B | 22 GB | Too big | no minimum listed |
| gpt-oss 20B | 20 GB | Too big | Ollama 0.30.0 or newer |
| Qwen 3.8 27B | 17.4 GB | Too big | Ollama 0.32.12 or newer |
| Gemma 4 26B | 19.2 GB | Too big | Ollama 0.30.9 or newer |
| Qwen 3 Coder 30B | 28 GB | Too big | no minimum listed |
| Qwen 3.5 35B | 35 GB | Too big | Ollama 0.30.0 or newer |
| Llama 3.3 70B | 60 GB | Too big | no minimum listed |
| gpt-oss 120B | 88 GB | Too big | Ollama 0.30.0 or newer |
If Ollama installs and runs, which is the untested part below, these start the two Qwen models rated for 8 GB.
# after installing Ollama from ollama.com
ollama run qwen3.5:4b # Qwen 3.5 4B
ollama run qwen3.5:2b # Qwen 3.5 2B, the smaller one
Each model's Ollama minimum is in the last column of the table. Larger Qwen, Gemma and gpt-oss models have their own pages: Qwen, Gemma and gpt-oss.
| Claim | Source |
|---|---|
| A18 Pro, 16-core Neural Engine, 8GB unified memory, 60GB/s, 256GB or 512GB SSD | Apple, MacBook Neo tech specs |
| Ollama: macOS 14 or newer, Apple M series (CPU and GPU support) or x86 (CPU only) | Ollama documentation, macOS |
| Grux AI: macOS 14 or later, Apple silicon | Grux README, Requirements |
| Every model's serving memory and the fit rule | Grux Cookbook.swift at the pinned commit |