Unified memory · Apple
What AI models can an Apple M3 Ultra 512GB run?
512 GB is server territory. The constraint stops being memory and becomes bandwidth: everything fits, and what varies is how long you wait for it. Memory is shared with the CPU, so the practical ceiling is lower than the sticker number: macOS hands roughly three quarters of it to the GPU by default.
The short answer
Assuming an 8k context window and default settings, these are the models worth downloading first.
See how fast it feels
Apple M3 Ultra 512GB running MiniMax M2.7 at Q8_0
YouWhy does my model use more memory when the conversation gets longer?
Reading your question
Simulated from our estimate at a 512-token question, not a recording. Assumes nothing else is competing for the GPU.
Where it sits
The GPU can use about 384 GB of this machine's 512 GB by default, so for planning it behaves like a 48 GB card with different bandwidth. The 48 GB guide explains what that tier runs and where it runs out, and the bandwidth guide explains why 819 GB/s decides how fast it answers.
Our indicative price for this machine is $9,499 new, reviewed 2026-09-01. That is 5.4 GB of memory and 9 GB/s of bandwidth for every $100. Prices move, so treat these as a place to start a search and not a quote.
Every model, scored on this device
Each row uses the highest-quality quantization that both fits and stays conversational. Speed is a single-stream estimate at 8k context.
| Model | Params | Verdict | Download | Memory used | Speed | Max context |
|---|---|---|---|---|---|---|
| MiniMax M2.7 MiniMax | 228.69B (10.98B active) | Runs great | Q8_0 | 228.58 GB | 25.3 t/s | 128k |
| MiniMax M2.5 MiniMax | 228.7B (10.98B active) | Runs great | Q8_0 | 228.59 GB | 25.3 t/s | 128k |
| Qwen3.5 122B A10B Qwen | 125.09B (8.17B active) | Runs great | Q8_0 | 124.45 GB | 38.6 t/s | 256k |
| GPT OSS 120B GPT-OSS | 116.83B (5.7B active) | Runs great | MXFP4 | 59.65 GB | 102.5 t/s | 128k |
| Qwen3 235B A22B Qwen | 235.09B (22.14B active) | Runs well | Q8_0 | 234.47 GB | 13.8 t/s | 32k |
| GLM 4.5 Air GLM | 110.47B (13.4B active) | Runs well | Q8_0 | 111.13 GB | 22.1 t/s | 128k |
| GLM 4.5 GLM | 358.34B (33.56B active) | Runs well | Q6_K | 276.94 GB | 11.4 t/s | 128k |
| Qwen3.5 27B Qwen | 27.78B | Runs well | Q8_0 | 28.54 GB | 20.7 t/s | 256k |
| Gemma 4 31B Instruct Gemma | 31.27B | Runs well | Q8_0 | 34.64 GB | 16.3 t/s | 256k |
| Qwen3 Coder Next Qwen | 79.67B (3.19B active) | Runs great | Q8_0 | 79.4 GB | 94.9 t/s | 256k |
| Qwen3 32B Qwen | 32.76B | Runs well | Q8_0 | 34.83 GB | 16.5 t/s | 32k |
| Qwen3 Next 80B A3B Instruct Qwen | 81.32B (3.19B active) | Runs great | Q8_0 | 81.04 GB | 94.9 t/s | 256k |
| Granite 4.1 30B Granite | 28.87B | Runs well | Q8_0 | 30.94 GB | 18.2 t/s | 128k |
| Qwen2.5 32B Instruct Qwen | 32.76B | Runs well | Q8_0 | 34.83 GB | 16.5 t/s | 32k |
| Gemma 3 27B Instruct Gemma | 27.43B | Runs well | Q8_0 | 28.65 GB | 19.7 t/s | 128k |
| Qwen2.5 Coder 32B Instruct Qwen | 32.76B | Runs well | Q8_0 | 34.83 GB | 16.5 t/s | 32k |
| Qwen3.5 35B A3B Qwen | 35.95B (2.9B active) | Runs great | Q8_0 | 36.1 GB | 106.1 t/s | 256k |
| Gemma 4 26B A4B Instruct Gemma | 25.81B (3.74B active) | Runs great | Q8_0 | 26.69 GB | 71.4 t/s | 256k |
| GLM 4.7 Flash GLM | 31.22B (3.66B active) | Runs great | Q8_0 | 31.61 GB | 80.8 t/s | 128k |
| Mixtral 8x7B Instruct V0.1 Mistral | 46.7B (12.88B active) | Runs well | Q8_0 | 47.59 GB | 23.6 t/s | 32k |
| Llama 3.3 70B Instruct Llama | 70.55B | Runs well | Q6_K | 56.89 GB | 10 t/s | 128k |
| Qwen2.5 72B Instruct Qwen | 72.71B | Runs well | Q5_K_M | 51.17 GB | 11.2 t/s | 32k |
| Qwen2.5 VL 72B Instruct Qwen | 73.41B | Runs well | Q5_K_M | 51.64 GB | 11.1 t/s | 64k |
| Qwen3 30B A3B Qwen | 30.53B (3.34B active) | Runs great | Q8_0 | 31.27 GB | 80.3 t/s | 32k |
| Mistral Small 24B Instruct 2501 Mistral | 23.57B | Runs well | Q8_0 | 24.98 GB | 23.3 t/s | 32k |
| Phi 3.5 MoE Instruct Phi | 41.87B (6.64B active) | Runs great | Q8_0 | 42.81 GB | 42.8 t/s | 128k |
| Gemma 4 12B Instruct Gemma | 11.96B | Runs great | Q8_0 | 13.52 GB | 42.4 t/s | 256k |
| Qwen3 14B Qwen | 14.77B | Runs great | Q8_0 | 16.28 GB | 36.9 t/s | 32k |
| GPT OSS 20B GPT-OSS | 20.91B (4.18B active) | Runs great | MXFP4 | 11.81 GB | 135.2 t/s | 128k |
| Qwen3 Coder 30B A3B Instruct Qwen | 30.53B (3.34B active) | Runs great | Q8_0 | 31.27 GB | 80.3 t/s | 32k |
Show the other 56 models
| Qwen2.5 14B Instruct | 14.77B | Runs great | Q8_0 | 16.53 GB | 36.3 t/s | 32k |
| Qwen2.5 Coder 14B Instruct | 14.77B | Runs great | Q8_0 | 16.53 GB | 36.3 t/s | 32k |
| Codestral 22B V0.1 | 22.25B | Runs well | Q8_0 | 24.21 GB | 23.6 t/s | 32k |
| Qwen3.5 9B | 9.65B | Runs great | Q8_0 | 10.22 GB | 63 t/s | 256k |
| Llama 3 3 Nemotron Super 49B V1 | 49.87B | Runs well | Q5_K_M | 53.55 GB | 10.6 t/s | 128k |
| Phi 4 | 14.66B | Runs great | Q8_0 | 16.48 GB | 35.8 t/s | 16k |
| Gemma 3 12B Instruct | 12.19B | Runs great | Q8_0 | 13.24 GB | 43.2 t/s | 128k |
| Gemma 4 E4B Instruct | 8B | Runs great | Q8_0 | 8.39 GB | 69.1 t/s | 128k |
| Granite 4.1 8B | 8.79B | Runs great | Q8_0 | 10.32 GB | 56 t/s | 128k |
| Qwen3 8B | 8.19B | Runs great | Q8_0 | 9.61 GB | 64.6 t/s | 32k |
| Mistral Nemo Instruct 2407 | 12.25B | Runs great | Q8_0 | 13.78 GB | 43.8 t/s | 128k |
| Qwen2.5 VL 7B Instruct | 8.29B | Runs great | Q8_0 | 9 GB | 68.7 t/s | 64k |
| Qwen2.5 7B Instruct | 7.62B | Runs great | Q8_0 | 8.34 GB | 74.9 t/s | 32k |
| Llama 3.1 8B Instruct | 8.03B | Runs great | Q8_0 | 9.32 GB | 66.1 t/s | 128k |
| DeepSeek Coder V2 Lite Instruct | 15.71B | Runs great | Q8_0 | 16.09 GB | 110.9 t/s | 128k |
| Gemma 2 9B Instruct | 9.24B | Runs great | Q8_0 | 11.47 GB | 50.1 t/s | 8k |
| Qwen2.5 Coder 7B Instruct | 7.62B | Runs great | Q8_0 | 8.34 GB | 74.9 t/s | 32k |
| OLMo 3 7B Instruct | 7.3B | Runs great | Q8_0 | 10.1 GB | 59.8 t/s | 64k |
| Internlm3 8B Instruct | 8.8B | Runs great | Q8_0 | 9.46 GB | 65 t/s | 32k |
| Mistral 7B Instruct V0.3 | 7.25B | Runs great | Q8_0 | 8.55 GB | 69.2 t/s | 32k |
| Gemma 4 E2B Instruct | 5.12B | Runs great | Q8_0 | 5.41 GB | 108.8 t/s | 128k |
| Qwen3.5 4B | 4.66B | Runs great | Q8_0 | 5.24 GB | 113.4 t/s | 256k |
| Mistral 7B Instruct V0.2 | 7.24B | Runs great | Q8_0 | 8.54 GB | 69.3 t/s | 32k |
| DeepSeek Coder 7B Instruct V1.5 | 6.91B | Runs great | Q8_0 | 10.96 GB | 54.7 t/s | 4k |
| DeepSeek Coder 6.7B Instruct | 6.74B | Runs great | Q8_0 | 11.04 GB | 52.8 t/s | 16k |
| LFM2.5 8B A1B | 8.47B | Runs great | Q8_0 | 8.78 GB | 196 t/s | 64k |
| Qwen3 4B | 4.02B | Runs great | Q8_0 | 5.43 GB | 109.1 t/s | 32k |
| Gemma 3 4B Instruct | 4.3B | Runs great | Q8_0 | 4.88 GB | 122.3 t/s | 128k |
| Phi 4 Mini Instruct | 3.84B | Runs great | Q8_0 | 5.14 GB | 116 t/s | 128k |
| Granite 4.1 3B | 3.4B | Runs great | Q8_0 | 4.31 GB | 139.6 t/s | 128k |
| Qwen2.5 3B Instruct | 3.09B | Runs great | Q8_0 | 3.65 GB | 166.8 t/s | 32k |
| Phi 3 Mini 4k Instruct | 3.82B | Runs great | Q8_0 | 4.87 GB | 125.6 t/s | 4k |
| Llama 3.2 3B Instruct | 3.21B | Runs great | Q8_0 | 4.39 GB | 137.4 t/s | 128k |
| LFM2.5 2.6B | 2.7B | Runs great | Q8_0 | 3.11 GB | 199.1 t/s | 128k |
| Qwen3.5 2B | 2.27B | Runs great | Q8_0 | 2.67 GB | 236.1 t/s | 256k |
| Gemma 2 2B Instruct | 2.61B | Runs great | Q8_0 | 3.51 GB | 174.4 t/s | 8k |
| Starcoder2 3B | 3.03B | Runs great | Q8_0 | 3.46 GB | 187.7 t/s | 16k |
| Qwen3 1.7B | 2.03B | Runs great | Q8_0 | 3.19 GB | 193.1 t/s | 32k |
| DeepSeek R1 Distill Qwen 1.5B | 1.78B | Runs great | Q8_0 | 2.27 GB | 318.3 t/s | 128k |
| Qwen2.5 1.5B Instruct | 1.54B | Runs great | Q8_0 | 2.03 GB | 319.5 t/s | 32k |
| SmolLM2 1.7B | 1.71B | Runs great | Q8_0 | 3.5 GB | 174.4 t/s | 8k |
| OLMo 2 0425 1B | 1.48B | Runs great | Q8_0 | 2.77 GB | 246.3 t/s | 4k |
| Llama 3.2 1B Instruct | 1.24B | Runs great | Q8_0 | 1.79 GB | 377 t/s | 128k |
| LFM2.5 1.2B Instruct | 1.17B | Runs great | Q8_0 | 1.56 GB | 497.7 t/s | 64k |
| Gemma 3 1B Instruct | 1B | Runs great | Q8_0 | 1.32 GB | 536 t/s | 32k |
| Qwen3.5 0.8B | 0.87B | Runs great | Q8_0 | 1.25 GB | 572.4 t/s | 256k |
| Qwen3 0.6B | 0.75B | Runs great | Q8_0 | 1.89 GB | 344.3 t/s | 32k |
| Qwen2.5 0.5B Instruct | 0.49B | Runs great | Q8_0 | 0.85 GB | 962.3 t/s | 32k |
| LFM2.5 350M | 0.35B | Runs great | Q8_0 | 0.71 GB | 1490.1 t/s | 64k |
| Kimi K2.6 | 1026.88B | Just fits | n/a | 468.43 GB | 17.7 t/s | n/a |
| Kimi K2 Instruct | 1026.41B | Just fits | n/a | 468.22 GB | 17.7 t/s | n/a |
| GLM 5.1 | 753.86B | Just fits | n/a | 425.01 GB | 15.5 t/s | n/a |
| DeepSeek V3.2 | 685.4B | Just fits | n/a | 455.02 GB | 12.4 t/s | n/a |
| DeepSeek R1 | 684.53B | Just fits | n/a | 454.45 GB | 12.4 t/s | n/a |
| MiniMax M3 | 427.04B | Just fits | n/a | 423.95 GB | 12.2 t/s | n/a |
| GLM 5.2 | 753.33B | Just fits | n/a | 424.72 GB | 10.9 t/s | n/a |
What it will not run
7 of the 86 architectures we track are out of reach here, even at 2-bit precision.
Questions people ask about an Apple M3 Ultra 512GB
How much VRAM does an Apple M3 Ultra 512GB have?
The Apple M3 Ultra 512GB has 512 GB of unified memory shared with the CPU, of which roughly 384 GB is available to a model once the operating system takes its share.
Can an Apple M3 Ultra 512GB run a 70B model?
Yes. At Q8_0 a 70B model fits in 79.4 GB and generates around 94.9 tokens per second.
What is the best local model for an Apple M3 Ultra 512GB?
MiniMax M2.7 at Q8_0 is the strongest all-round fit: 228.58 GB used, around 25.3 tokens per second at 8k context.