What Can the Apple M5 Ultra Run?
96 GB · compatibility with 168 AI models
Short answer: the Apple M5 Ultra (96 GB) can run 168 of 168 AI image, video, 3D and audio models we track. Best picks: SDXL Base 1.0, SDXL img2img, Stable Diffusion 1.5, SD 1.5 img2img, SD 1.5 Inpainting. Unified memory, up to 512GB config.
| Perfect | Great | Good | Usable | Barely | Can't Run |
|---|---|---|---|---|---|
| 13 models | 36 models | 0 models | 119 models | 0 models | 0 models |
Apple M5 Ultra specs for AI
| Brand | Apple |
|---|---|
| Series | M5 |
| VRAM | 96 GB |
| Architecture | m5 |
| FP16 compute | 27.2 TFLOPS |
| Memory bandwidth | 1228 GB/s |
| Released | 2026 |
| Launch price | $5499 |
AI models on the Apple M5 Ultra
Each model uses the best format that fits in 96 GB. How ratings work.
Text to Image
| Model | Rating | Runs as | VRAM used | Speed |
|---|---|---|---|---|
| SDXL Base 1.0 | Great | FP16 | 7 GB | Fast |
| Stable Diffusion 1.5 | Great | FP16 | 4 GB | Fast |
| SDXL Turbo | Great | FP16 | 7 GB | Fast |
| Stable Diffusion 1.4 | Great | FP16 | 4 GB | Fast |
| Stable Diffusion 3.5 Medium (2.5B) | Great | FP16 | 8 GB | Fast |
| Animagine XL 3.1 | Great | FP16 | 7 GB | Fast |
| Pony Diffusion v6 XL | Great | FP16 | 7 GB | Fast |
| NoobAI XL | Great | FP16 | 7 GB | Fast |
| Juggernaut XL Ragnarok | Great | FP16 | 7 GB | Fast |
| SDXL Lightning 4-step | Great | FP16 | 7 GB | Fast |
| DreamShaper XL | Great | FP16 | 7 GB | Fast |
| Illustrious XL v1.0 | Great | FP16 | 7 GB | Fast |
| Playground v2.5 2.6B | Great | FP16 | 7 GB | Fast |
| RealVisXL v5.0 | Great | FP16 | 7 GB | Fast |
| Epicrealism XL | Great | FP16 | 7 GB | Fast |
| PixArt-Sigma 900M | Great | FP16 | 7 GB | Fast |
| Stable Diffusion 2.1 | Great | FP16 | 5 GB | Fast |
| PixArt-Alpha 600M | Great | FP16 | 6 GB | Fast |
| Stable Diffusion 3 Medium (2B) | Great | FP16 | 7 GB | Fast |
| SDXL Hyper 1-step | Great | FP16 | 7 GB | Fast |
| Stable Diffusion 2.0 | Great | FP16 | 5 GB | Fast |
| Z-Image Turbo | Usable | FP16 | 10 GB | Fast |
| HunyuanDiT 1.5B | Usable | FP16 | 8 GB | Fast |
| Qwen-Image 2.1 | Usable | FP16 | 14 GB | Fast |
| MOSS-Image 5B | Usable | FP16 | 12 GB | Fast |
| Flux.2 Klein 4B | Usable | FP16 | 13 GB | Fast |
| Z-Image Base | Usable | FP16 | 14 GB | Fast |
| Ming-Image 0.1 Design | Usable | FP16 | 14 GB | Fast |
| ERNIE-Image 8B | Usable | FP16 | 16 GB | Fast |
| HiDream-O1-Image 8.8B | Usable | FP16 | 16 GB | Fast |
| Seedream 3.0 | Usable | FP16 | 16 GB | Fast |
| GLM-Image | Usable | FP16 | 16 GB | Fast |
| Flux.1 Lite | Usable | FP16 | 12 GB | Fast |
| AuraFlow 0.3 | Usable | FP16 | 12 GB | Fast |
| Kolors | Usable | FP16 | 14 GB | Fast |
| Flux.1 Dev 12B | Usable | FP16 | 24 GB | Fast |
| FLUX.2 Klein 9B | Usable | FP16 | 18 GB | Fast |
| Flux.1 Schnell 12B | Usable | FP16 | 24 GB | Fast |
| Stable Diffusion 3.5 Large (8B) | Usable | FP16 | 18 GB | Fast |
| Stable Diffusion 3.5 Large Turbo | Usable | FP16 | 18 GB | Fast |
| Ideogram 4.0 | Usable | FP16 | 19 GB | Fast |
| Chroma1 HD | Usable | FP16 | 18 GB | Fast |
| FLUX.1 Krea Dev | Usable | FP16 | 24 GB | Fast |
| HunyuanImage 3.0 | Usable | FP16 | 20 GB | Fast |
| Krea 2 Turbo | Usable | FP16 | 26 GB | Fast |
| Qwen-Image 2512 (20B) | Usable | FP16 | 41 GB | Fast |
| Flux.2 Dev 32B | Usable | FP16 | 64 GB | Fast |
Image to Image
| Model | Rating | Runs as | VRAM used | Speed |
|---|---|---|---|---|
| SDXL img2img | Great | FP16 | 7 GB | Fast |
| SD 1.5 img2img | Great | FP16 | 4 GB | Fast |
| InstructPix2Pix | Great | FP16 | 5 GB | Fast |
| IP-Adapter FaceID SDXL | Usable | FP16 | 10 GB | Fast |
| IP-Adapter SDXL | Usable | FP16 | 9 GB | Fast |
| Mage-Flow 4B | Usable | FP16 | 10 GB | Fast |
| LongCat-Image Edit | Usable | FP16 | 15 GB | Fast |
| Flux.1 Dev img2img 12B | Usable | FP16 | 24 GB | Fast |
| Flux Kontext Dev | Usable | FP16 | 24 GB | Fast |
| Boogu-Image Edit | Usable | FP16 | 21 GB | Fast |
| Qwen-Image-Edit 2511 | Usable | FP16 | 41 GB | Fast |
Inpainting
| Model | Rating | Runs as | VRAM used | Speed |
|---|---|---|---|---|
| SD 1.5 Inpainting | Great | FP16 | 4 GB | Fast |
| SDXL Inpainting | Great | FP16 | 7 GB | Fast |
| BrushNet | Usable | FP16 | 8 GB | Fast |
| PowerPaint v2.1 | Usable | FP16 | 7 GB | Fast |
| Flux.1 Fill Dev | Usable | FP16 | 24 GB | Fast |
ControlNet
| Model | Rating | Runs as | VRAM used | Speed |
|---|---|---|---|---|
| ControlNet SD 1.5 | Great | FP16 | 5 GB | Fast |
| ControlNet SDXL | Usable | FP16 | 9 GB | Fast |
| T2I-Adapter SDXL | Usable | FP16 | 8 GB | Fast |
| ControlNet Flux | Usable | FP16 | 28 GB | Fast |
Video
| Model | Rating | Runs as | VRAM used | Speed |
|---|---|---|---|---|
| WAN 2.1 T2V 1.3B | Usable | FP16 | 6 GB | Fast |
| LTX-Video 0.9 (2B) | Usable | FP16 | 10 GB | Fast |
| LTX-Video I2V | Usable | FP16 | 10 GB | Fast |
| Stable Video Diffusion 1.1 | Usable | FP16 | 8 GB | Fast |
| AnimateDiff Lightning | Usable | FP16 | 6 GB | Fast |
| CogVideoX 2B | Usable | FP16 | 8 GB | Fast |
| AnimateDiff v3 (SD 1.5) | Usable | FP16 | 6 GB | Fast |
| ModelScope T2V | Usable | FP16 | 8 GB | Fast |
| Stable Video Diffusion XT 1.1 (25 frames) | Usable | FP16 | 10 GB | Fast |
| ZeroScope v2 576w | Usable | FP16 | 6 GB | Fast |
| DynamiCrafter 512 | Usable | FP16 | 10 GB | Fast |
| TokenFlow | Usable | FP16 | 8 GB | Fast |
| AnimateDiff V2V | Usable | FP16 | 6 GB | Fast |
| Latte | Usable | FP16 | 8 GB | Fast |
| AnimateDiff I2V | Usable | FP16 | 6 GB | Fast |
| SEINE | Usable | FP16 | 8 GB | Fast |
| Kandinsky 6.0 Video Lite 3B | Usable | FP16 | 9 GB | Fast |
| WAN 2.2 TI2V 5B | Usable | FP16 | 12 GB | Fast |
| CogVideoX 5B | Usable | FP16 | 14 GB | Fast |
| CogVideoX 5B I2V | Usable | FP16 | 14 GB | Fast |
| SkyReels V1 | Usable | FP16 | 14 GB | Fast |
| SkyReels V1 I2V | Usable | FP16 | 14 GB | Fast |
| AnimateDiff SDXL | Usable | FP16 | 12 GB | Fast |
| Waver 1.0 | Usable | FP16 | 12 GB | Fast |
| Open-Sora 2.0 (11B) | Usable | FP16 | 24 GB | Fast |
| MAGI-1 | Usable | FP16 | 24 GB | Fast |
| DynamiCrafter 1024 | Usable | FP16 | 18 GB | Fast |
| Ovi 1.1 | Usable | FP16 | 23 GB | Fast |
| WAN 2.2 T2V A14B (MoE) | Usable | FP16 | 28 GB | Fast |
| WAN 2.2 I2V A14B | Usable | FP16 | 28 GB | Fast |
| WAN 2.1 T2V 14B | Usable | FP16 | 28 GB | Fast |
| WAN 2.1 I2V 480P 14B | Usable | FP16 | 28 GB | Fast |
| WAN 2.1 V2V | Usable | FP16 | 28 GB | Fast |
| HunyuanVideo T2V v1.5 | Usable | FP16 | 28 GB | Fast |
| HunyuanVideo I2V v1.5 | Usable | FP16 | 28 GB | Fast |
| HunyuanVideo 13B T2V | Usable | FP16 | 28 GB | Fast |
| HunyuanVideo I2V | Usable | FP16 | 28 GB | Fast |
| LTX-Video 13B | Usable | FP16 | 28 GB | Fast |
| MOVA | Usable | FP16 | 29 GB | Fast |
| LongCat-Video 13.6B | Usable | FP16 | 27 GB | Fast |
| SkyReels V3 R2V 14B | Usable | FP16 | 29 GB | Fast |
| Wan 2.2 Animate 14B | Usable | FP16 | 35 GB | Fast |
| Wan 2.2 S2V 14B | Usable | FP16 | 33 GB | Fast |
| WAN 2.1 I2V 720P 14B | Usable | FP16 | 32 GB | Fast |
| Wan-Animate-2 14B | Usable | FP16 | 33 GB | Fast |
| SCAIL-2 14B | Usable | FP16 | 33 GB | Fast |
| FramePack I2V HY FP8 | Usable | FP16 | 32 GB | Fast |
| Wan-Dancer 14B | Usable | FP16 | 34 GB | Fast |
| HuMo 17B | Usable | FP16 | 34 GB | Fast |
| MiniMax H3 FL2VA | Usable | FP16 | 40 GB | Fast |
| MiniMax H3 Ref2VA | Usable | FP16 | 40 GB | Fast |
| LTX-2.5 22B | Usable | FP16 | 42 GB | Fast |
| LTX-2.3 22B | Usable | FP16 | 42 GB | Fast |
| LTX-2 19B | Usable | FP16 | 38 GB | Fast |
| Mochi 1 10B | Usable | FP16 | 40 GB | Fast |
Upscaler
| Model | Rating | Runs as | VRAM used | Speed |
|---|---|---|---|---|
| 4x-UltraSharp | Perfect | FP32 | 0.3 GB | Fast |
| 4x-Remacri | Perfect | FP32 | 0.3 GB | Fast |
| SwinIR | Perfect | FP32 | 0.5 GB | Fast |
| RealESRGAN x4plus | Perfect | FP32 | 0.3 GB | Fast |
| HAT | Perfect | FP32 | 0.5 GB | Fast |
| DAT | Perfect | FP32 | 0.5 GB | Fast |
| ScuNET | Perfect | FP32 | 0.3 GB | Fast |
| LDSR | Great | FP16 | 4 GB | Fast |
| SeedVR2 3B | Usable | FP16 | 9 GB | Fast |
| SeedVR2 7B | Usable | FP16 | 16 GB | Fast |
Face/Restore
| Model | Rating | Runs as | VRAM used | Speed |
|---|---|---|---|---|
| GFPGAN v1.4 | Perfect | FP32 | 0.3 GB | Fast |
| CodeFormer v0.1 | Perfect | FP32 | 0.3 GB | Fast |
VAE
| Model | Rating | Runs as | VRAM used | Speed |
|---|---|---|---|---|
| SD 1.5 VAE ft-mse | Perfect | FP32 | 0.2 GB | Fast |
| Flux VAE | Perfect | FP32 | 0.3 GB | Fast |
| SDXL VAE | Perfect | FP32 | 0.3 GB | Fast |
| TAESD / TAESDXL | Perfect | FP32 | 0.1 GB | Fast |
3D
| Model | Rating | Runs as | VRAM used | Speed |
|---|---|---|---|---|
| TripoSR | Great | FP16 | 4 GB | Fast |
| Shap-E | Great | FP16 | 4 GB | Fast |
| Hunyuan3D 2.1 | Usable | FP16 | 10 GB | Fast |
| InstantMesh | Usable | FP16 | 6 GB | Fast |
| SV3D | Usable | FP16 | 10 GB | Fast |
| Zero123++ v1.2 | Usable | FP16 | 6 GB | Fast |
| LGM | Usable | FP16 | 8 GB | Fast |
| One-2-3-45 | Usable | FP16 | 8 GB | Fast |
| Pixal3D | Usable | FP16 | 14 GB | Fast |
| Hunyuan3D 2.0 | Usable | FP16 | 12 GB | Fast |
| TRELLIS.2 4B | Usable | FP16 | 24 GB | Fast |
| SAM 3D Objects | Usable | FP16 | 32 GB | Fast |
Audio
| Model | Rating | Runs as | VRAM used | Speed |
|---|---|---|---|---|
| MusicGen Small (300M) | Great | FP16 | 2 GB | Fast |
| Whisper Large v3 | Great | FP16 | 4 GB | Fast |
| MusicGen Medium (1.5B) | Great | FP16 | 6 GB | Fast |
| Bark | Great | FP16 | 8 GB | Fast |
| XTTS v2 | Great | FP16 | 3 GB | Fast |
| Stable Audio Open | Great | FP16 | 4 GB | Fast |
| Qwen3-TTS 1.7B | Usable | FP16 | 5 GB | Fast |
| Chatterbox Multilingual | Usable | FP16 | 4 GB | Fast |
| ACE-Step 1.5 Turbo | Usable | FP16 | 8 GB | Fast |
| VibeVoice 1.5B | Usable | FP16 | 7 GB | Fast |
| Stable Audio 3 Medium | Usable | FP16 | 7 GB | Fast |
| Stable Audio 3 Small-SFX | Usable | FP16 | 3 GB | Fast |
| MMAudio Large 44k | Usable | FP16 | 4 GB | Fast |
| IndexTTS 2.5 | Usable | FP16 | 6 GB | Fast |
| MusicGen Large (3.3B) | Usable | FP16 | 12 GB | Fast |
| YuE2 3B | Usable | FP16 | 24 GB | Fast |
| MiniMax Music 3 | Usable | FP16 | 24 GB | Fast |
| Fish Audio S2 Pro 4B | Usable | FP16 | 24 GB | Fast |
Frequently asked questions
How much VRAM does the Apple M5 Ultra have?
The Apple M5 Ultra has 96 GB of unified memory (shared with the CPU) with 1228 GB/s of memory bandwidth.
What AI models can the Apple M5 Ultra run?
It can run 168 of the 168 AI models we track: 13 perfectly (S), 36 great (A), 0 good (B), 119 usable (C) and 0 barely (D). 0 need more VRAM than it has.
Can the Apple M5 Ultra run MiniMax H3 FL2VA?
Yes, in FP16 (rating C, usable).
Can the Apple M5 Ultra run MiniMax H3 Ref2VA?
Yes, in FP16 (rating C, usable).
Can the Apple M5 Ultra run Qwen-Image 2.1?
Yes, in FP16 (rating C, usable).
Can the Apple M5 Ultra run Z-Image Turbo?
Yes, in FP16 (rating C, usable).