# Can the Apple M4 Pro (48 GB) run Axion-1.0-35B?

> Yes. The Apple M4 Pro (48 GB) runs Axion-1.0-35B fully in unified memory at Q6_K (about 29 GB of 36 GB), at roughly 47 tokens/s.

Apple M4 Pro (48 GB): 48 GB (36 GB usable), 273 GB/s. Estimates at an 8K context.

## Every quantisation

| Quant | Quality | ≈ GB | Fit | Est. tok/s |
|---|---|---|---|---|
| Q2_K | Heavy loss | 13.78 | fits | 117 |
| Q3_K_M | Noticeable loss | 17.52 | fits | 79 |
| Q4_K_M | Small loss | 21.83 | fits | 63 |
| Q5_K_M | Near-lossless | 25.31 | fits | 54 |
| Q6_K (pick) | Near-lossless | 29.02 | fits | 47 |
| Q8_0 | Lossless | 37.22 | no | — |
| F16 | Lossless | 69.79 | no | — |

## Run it

Apple Silicon with mlx-lm (mlx):

```bash
pip install mlx-lm
mlx_lm.generate --model NapYang/Axion-1.0-35B-UD-MLX-4bit-fp16 --prompt "Hello"
```

OpenAI-compatible server (mlx):

```bash
mlx_lm.server --model NapYang/Axion-1.0-35B-UD-MLX-4bit-fp16 --port 8080
```

Q6_K straight from Hugging Face (ollama):

```bash
ollama run hf.co/mradermacher/Axion-1.0-35B-GGUF:Q6_K
```

OpenAI-compatible server with a web UI (llama.cpp):

```bash
llama-server -hf mradermacher/Axion-1.0-35B-GGUF:Q6_K
```

Download with the lms CLI (lmstudio):

```bash
lms get mradermacher/Axion-1.0-35B-GGUF@Q6_K
```

Chat with the pipeline API (transformers):

```python
from transformers import pipeline

pipe = pipeline("text-generation", model="Axion1-0/Axion-1.0-35B", dtype="auto", device_map="auto")
messages = [{"role": "user", "content": "Explain self-hosting in one paragraph."}]
out = pipe(messages, max_new_tokens=512)
print(out[0]["generated_text"][-1]["content"])
```


## Chat models that fit comfortably

- [ornith-ai/Ornith-1.5-35B-A3B](https://selfhost.directory/models/ornith-ai/ornith-1.5-35b-a3b/on/apple-m4-pro-48gb): Q6_K ≈30.98 GB, est. 61 tok/s
- [zai-org/GLM-4.7-Flash](https://selfhost.directory/models/zai-org/glm-4.7-flash/on/apple-m4-pro-48gb): Q6_K ≈31.99 GB, est. 40 tok/s
- [Qwen/Qwen3-32B](https://selfhost.directory/models/qwen/qwen3-32b/on/apple-m4-pro-48gb): Q6_K ≈28.79 GB, est. 5.6 tok/s
- [RecursiveSI/flint](https://selfhost.directory/models/recursivesi/flint/on/apple-m4-pro-48gb): Q6_K ≈33.31 GB, est. 146 tok/s
- [openai/gpt-oss-20b](https://selfhost.directory/models/openai/gpt-oss-20b/on/apple-m4-pro-48gb): Q8_0 ≈12.72 GB, est. 34 tok/s
- [Qwen/Qwen3-30B-A3B](https://selfhost.directory/models/qwen/qwen3-30b-a3b/on/apple-m4-pro-48gb): Q8_0 ≈33.01 GB, est. 47 tok/s

## Other GPUs that run it

- [Tesla P40](https://selfhost.directory/models/axion1-0/axion-1.0-35b/on/tesla-p40): Q4_K_M, est. 81 tok/s
- [Radeon RX 7900 XTX](https://selfhost.directory/models/axion1-0/axion-1.0-35b/on/rx-7900-xtx): Q4_K_M, est. 223 tok/s
- [Instinct MI50 32GB](https://selfhost.directory/models/axion1-0/axion-1.0-35b/on/instinct-mi50): Q6_K, est. 175 tok/s
- [GeForce RTX 5090](https://selfhost.directory/models/axion1-0/axion-1.0-35b/on/rtx-5090): Q6_K, est. 307 tok/s
- [RTX A6000](https://selfhost.directory/models/axion1-0/axion-1.0-35b/on/rtx-a6000): Q8_0, est. 102 tok/s
- [Apple M1 Max (64 GB)](https://selfhost.directory/models/axion1-0/axion-1.0-35b/on/apple-m1-max-64gb): Q8_0, est. 53 tok/s
- [NVIDIA A100 80GB](https://selfhost.directory/models/axion1-0/axion-1.0-35b/on/a100-80gb): F16, est. 144 tok/s
- [Apple M4 Max 40-core (128 GB)](https://selfhost.directory/models/axion1-0/axion-1.0-35b/on/apple-m4-max-128gb-40c): F16, est. 38 tok/s

---
Model: https://selfhost.directory/models/axion1-0/axion-1.0-35b. GPU: https://selfhost.directory/models/gpus/apple-m4-pro-48gb. Source: https://selfhost.directory/models/axion1-0/axion-1.0-35b/on/apple-m4-pro-48gb