Apple M3 Ultra vs Apple M4 Max
Head-to-head AI inference comparison across 109 popular models. Each model is graded against both cards using its highest-quality quantization that still fits in VRAM. Bigger grade and faster tokens-per-second wins.
Spec
- VRAM
- Architecture
- Vendor
- MSRP
- Models running
- Wins (grade)
Apple M3 Ultra
- 192GB
- m3
- apple
- —
- 109 of 109
- 0 models
Apple M4 Max
- 128GB
- m4
- apple
- —
- 109 of 109
- 0 models
Language Models47 models
Ultra
tok/s
Model
tok/s
Max
S
22
Llama 3.1 70B Instruct
70B · Meta
17
S
S
30
Qwen 2.5 32B
32B · Alibaba
26
S
S
30
Gemma 3 27B
27B · Google
26
S
S
30
Mistral Small 22B
22B · Mistral AI
26
S
S
42
Phi-4
14B · Microsoft
36
S
S
42
Qwen 2.5 14B
14B · Alibaba
36
S
S
42
Gemma 3 12B
12B · Google
36
S
S
42
Mistral Nemo 12B
12B · Mistral AI
36
S
S
42
Solar 10.7B
10.7B · Upstage
36
S
S
54
Falcon 3 10B
10B · TII
48
S
S
54
Gemma 2 9B Instruct
9.2B · Google
48
S
S
54
Yi 1.5 9B Chat
9B · 01.AI
48
S
S
54
DeepSeek R1 Distill 8B
8B · DeepSeek
48
S
S
54
Llama 3.1 8B Instruct
8B · Meta
48
S
S
54
Granite 3.3 8B
8B · IBM
48
S
S
54
EXAONE 3.5 7.8B
7.8B · LG AI
48
S
S
54
InternLM 2.5 7B
7.7B · Shanghai AI Lab
48
S
S
54
Qwen 2.5 7B Instruct
7.6B · Alibaba
48
S
S
54
Mistral 7B Instruct v0.3
7.3B · Mistral AI
48
S
S
54
Falcon 3 7B
7B · TII
48
S
S
54
OLMo 2 7B
7B · Allen AI
48
S
S
54
OpenChat 3.5 7B
7B · OpenChat
48
S
S
54
Yi 1.5 6B Chat
6B · 01.AI
48
S
S
84
Gemma 3 4B
4B · Google
74
S
S
84
Nemotron Mini 4B
4B · NVIDIA
74
S
S
84
Danube 3 4B
4B · H2O.ai
74
S
S
84
Phi-3.5 Mini 3.8B
3.8B · Microsoft
74
S
S
84
Phi-4 Mini 3.8B
3.8B · Microsoft
74
S
S
84
Llama 3.2 3B Instruct
3.2B · Meta
74
S
S
84
Qwen 2.5 3B
3B · Alibaba
74
S
S
84
Falcon 3 3B
3B · TII
74
S
S
84
StableLM Zephyr 3B
3B · Stability AI
74
S
S
84
Rocket 3B
3B · Pansophic
74
S
S
84
Gemma 2 2B
2.6B · Google
74
S
S
84
EXAONE 3.5 2.4B
2.4B · LG AI
74
S
S
114
Granite 3.3 2B
2B · IBM
102
S
S
114
SmolLM2 1.7B
1.7B · HuggingFace
102
S
S
114
Qwen 2.5 1.5B
1.5B · Alibaba
102
S
S
114
DeepSeek R1 Distill 1.5B
1.5B · DeepSeek
102
S
S
114
Llama 3.2 1B Instruct
1.24B · Meta
102
S
S
114
TinyLlama 1.1B
1.1B · TinyLlama
102
S
S
114
Gemma 3 1B
1B · Google
102
S
S
114
Falcon 3 1B
1B · TII
102
S
S
114
Qwen 2.5 0.5B
0.5B · Alibaba
102
S
S
114
Danube 3 500M
0.5B · H2O.ai
102
S
S
114
SmolLM2 360M
0.36B · HuggingFace
102
S
S
114
SmolLM2 135M
0.135B · HuggingFace
102
S
Code Models16 models
Ultra
tok/s
Model
tok/s
Max
S
42
Qwen 2.5 Coder 14B
14B · Alibaba
36
S
S
42
Code Llama 13B Instruct
13B · Meta
36
S
S
54
Yi Coder 9B
9B · 01.AI
48
S
S
54
CodeGemma 7B
8.5B · Google
48
S
S
54
Qwen 2.5 Coder 7B
7.6B · Alibaba
48
S
S
54
StarCoder2 7B
7B · BigCode
48
S
S
54
Code Llama 7B
7B · Meta
48
S
S
54
DeepSeek Coder 6.7B
6.7B · DeepSeek
48
S
S
84
Qwen 2.5 Coder 3B
3B · Alibaba
74
S
S
84
StarCoder2 3B
3B · BigCode
74
S
S
84
Stable Code 3B
3B · Stability AI
74
S
S
114
CodeGemma 2B
2B · Google
102
S
S
114
Qwen 2.5 Coder 1.5B
1.5B · Alibaba
102
S
S
114
Yi Coder 1.5B
1.5B · 01.AI
102
S
S
114
DeepSeek Coder 1.3B
1.3B · DeepSeek
102
S
S
114
Qwen 2.5 Coder 0.5B
0.5B · Alibaba
102
S
Multimodal & Vision6 models
Ultra
tok/s
Model
tok/s
Max
Image Generation9 models
Ultra
tok/s
Model
tok/s
Max
S
42
FLUX.1 Schnell (GGUF)
12B · Black Forest Labs
36
S
S
42
FLUX.1 Dev (GGUF)
12B · Black Forest Labs
36
S
S
84
Stable Diffusion XL (CoreML)
3.5B · Stability AI
74
S
S
84
SDXL Turbo (GGUF)
3.5B · Stability AI
74
S
S
84
Stable Diffusion 3 Medium (GGUF)
2.5B · Stability AI
74
S
S
114
Stable Diffusion 2.1 Base (CoreML)
0.86B · Stability AI / Apple
102
S
S
114
Stable Diffusion 1.5 (CoreML)
0.86B · Runway
102
S
S
114
Stable Diffusion 1.5 (GGUF)
0.86B · Runway / GPUStack
102
S
S
114
Stable Diffusion 2.1 (GGUF)
0.86B · Stability AI
102
S
Speech9 models
Ultra
tok/s
Model
tok/s
Max
S
114
Whisper Large v3
1.55B · OpenAI
102
S
S
114
Whisper Large v3 Turbo
0.81B · OpenAI
102
S
S
114
Whisper Medium
0.77B · OpenAI
102
S
S
114
Distil-Whisper Large v3
0.76B · HuggingFace
102
S
S
114
Whisper Small
0.24B · OpenAI
102
S
S
114
Whisper Base
0.074B · OpenAI
102
S
S
114
Whisper Base English
0.074B · OpenAI
102
S
S
114
Whisper Tiny English (Quantized)
0.039B · OpenAI
102
S
S
114
Whisper Tiny
0.039B · OpenAI
102
S
Text-to-Speech14 models
Ultra
tok/s
Model
tok/s
Max
S
114
Kokoro 82M TTS
0.082B · Kokoro
102
S
S
114
Piper TTS - Amy (English)
0.02B · Rhasspy
102
S
S
114
Piper TTS - Lessac (English)
0.02B · Rhasspy
102
S
S
114
Piper TTS - LibriTTS-R (English)
0.02B · Rhasspy
102
S
S
114
Piper TTS - Spanish (MLS)
0.02B · Rhasspy
102
S
S
114
Piper TTS - French (Siwis)
0.02B · Rhasspy
102
S
S
114
Piper TTS - German (Thorsten)
0.02B · Rhasspy
102
S
S
114
Piper TTS - Chinese (Huayan)
0.02B · Rhasspy
102
S
S
114
Piper TTS - Japanese (Kokoro)
0.02B · Rhasspy
102
S
S
114
Piper TTS - Korean
0.02B · Rhasspy
102
S
S
114
Piper TTS - Russian (Irina)
0.02B · Rhasspy
102
S
S
114
Piper TTS - Portuguese (Faber)
0.02B · Rhasspy
102
S
S
114
Piper TTS - Italian (Riccardo)
0.02B · Rhasspy
102
S
S
114
Piper TTS - Arabic (Kareem)
0.02B · Rhasspy
102
S
Embeddings5 models
Ultra
tok/s
Model
tok/s
Max
Rerankers2 models
Ultra
tok/s
Model
tok/s
Max