Apple M5 Pro
LaptopApple unified-memory SoC, launched with the 2026-03 MacBook Pro refresh: up to 64 GB of unified memory at 307 GB/s, built on Apple's new Fusion architecture. Apple does not publish a TDP, so that column is left empty.
43
Run records
18
Models covered
4
Frameworks covered
L0 Self-reported × 7 · L2 Cross-framework verified × 36
Evidence mix
Typical measured performance
Published benchmark averages per model × framework × quantization on this hardware.
- Ollama · Q5_K_M Decode 121.74 · Prefill 552.05 ·
- Ollama · Q6_K Decode 111.41 · Prefill 526.06 ·
- vLLM · Q3_K_L Decode 145.87 · Prefill 596.68 ·
- vLLM · Q3_K_S Decode 149.44 · Prefill 576.02 ·
- llama.cpp · IQ1_S Decode 341.6 · Prefill 1406.03 ·
- vLLM · IQ2_M Decode 323.11 · Prefill 1474.14 ·
- vLLM · Q3_K_M Decode 358.52 · Prefill 1906.72 ·
- vLLM · IQ3_S Decode 158.08 · Prefill 657.93 ·
- vLLM · Q2_K Decode 169.33 · Prefill 690.81 ·
- Ollama · IQ3_S Decode 136.51 · Prefill 597.98 ·
- Ollama · Q3_K_S Decode 144.08 · Prefill 671.77 ·
- Ollama · IQ1_S Decode 187.29 · Prefill 770.8 ·
- Ollama · Q4_K_S Decode 116.66 · Prefill 579.89 ·
- Strata · Q3_K_L Decode 212.69 · Prefill 1110.56 ·
- llama.cpp · Q3_K_L Decode 189.13 · Prefill 672.33 ·
- llama.cpp · Q3_K_M Decode 187.03 · Prefill 648.58 ·
- llama.cpp · IQ2_M Decode 311.55 · Prefill 1641.2 ·
- llama.cpp · Q5_K_S Decode 197.87 · Prefill 839.49 ·
- vLLM · Q4_K_M Decode 233.37 · Prefill 1190.9 ·
- vLLM · Q5_K_S Decode 229.77 · Prefill 936.19 ·
- llama.cpp · Q3_K_L Decode 181 · Prefill 762.88 ·
- vLLM · Q5_K_S Decode 376.55 · Prefill 1237.86 ·
- vLLM · Q8_0 Decode 284.14 · Prefill 1555.29 ·
- Ollama · Q5_K_M Decode 149.09 · Prefill 749.3 ·
- llama.cpp · IQ1_S Decode 286.68 · Prefill 1382.17 ·
- llama.cpp · IQ2_M Decode 301.75 · Prefill 1278.13 ·
- Ollama · Q4_K_S Decode 342.8 · Prefill 1259.06 ·
- Ollama · Q3_K_M Decode 62.31 · Prefill 340.97 ·
- Strata · IQ3_XS Decode 111.36 · Prefill 546.98 ·
- llama.cpp · Q8_0 Decode 55.26 · Prefill 261.28 ·
- Strata · Q6_K Decode 98.9 · Prefill 379.8 ·
- llama.cpp · Q4_K_M Decode 100.37 · Prefill 374.04 ·
- llama.cpp · Q8_0 Decode 66.87 · Prefill 301.78 ·
- Ollama · Q3_K_M Decode 110.36 · Prefill 569.49 ·
- Ollama · Q3_K_S Decode 111.33 · Prefill 497.48 ·
- llama.cpp · Q2_K Decode 127.36 · Prefill 409.89 ·
- llama.cpp · IQ2_XS Decode 0.9 · Prefill 6.8 ·
- llama.cpp · IQ3_S Decode 0.8 · Prefill 6 ·
- llama.cpp · IQ2_XS Decode 0.9 · Prefill 6.8 ·
- llama.cpp · IQ3_S Decode 0.8 · Prefill 6 ·
- llama.cpp · IQ2_XS Decode 26.3 · Prefill 197.3 ·
- llama.cpp · Q5_K_M Decode 12 · Prefill 90 ·
- llama.cpp · Q4_K_M Decode 14.6 · Prefill 109.5 ·
Fit check
Physics-based estimate from VRAM and memory bandwidth: pick a model to see which quantizations fit and the expected speed
Formula: tok/s = (bandwidth ÷ model size) × 0.55, adapted from llmfit; see run records for measured numbers
Key specs
- Vendor
- Apple
- Architecture
- M5 Pro
- VRAM
- 64 GB
- Memory bandwidth
- 307 GB/s
Run records
43 configurations
| Model | Framework | Quantization | Decode | Gen time | Samples |
|---|---|---|---|---|---|
| Qwen3.5-4B | Ollama | Q5_K_M | 121.7 tok/s | — | 1 |
| Qwen3.5-4B | Ollama | Q6_K | 111.4 tok/s | — | 1 |
| Qwen3.5-9B | vLLM | Q3_K_L | 145.9 tok/s | — | 1 |
| Qwen3.5-9B | vLLM | Q3_K_S | 149.4 tok/s | — | 1 |
| Qwen3.5-0.8B | llama.cpp | IQ1_S | 341.6 tok/s | — | 1 |
| Qwen3.5-0.8B | vLLM | IQ2_M | 323.1 tok/s | — | 1 |
| Qwen3.5-0.8B | vLLM | Q3_K_M | 358.5 tok/s | — | 1 |
| Qwen3.8-9B | vLLM | IQ3_S | 158.1 tok/s | — | 1 |
| Qwen3.8-9B | vLLM | Q2_K | 169.3 tok/s | — | 1 |
| Qwen3.8-9B | Ollama | IQ3_S | 136.5 tok/s | — | 1 |
| Qwen3.8-9B | Ollama | Q3_K_S | 144.1 tok/s | — | 1 |
| Qwen3-8B | Ollama | IQ1_S | 187.3 tok/s | — | 1 |
| Qwen3-8B | Ollama | Q4_K_S | 116.7 tok/s | — | 1 |
| Qwen3-8B | Strata | Q3_K_L | 212.7 tok/s | — | 1 |
| Qwen3-4B | llama.cpp | Q3_K_L | 189.1 tok/s | — | 1 |
| Qwen3-4B | llama.cpp | Q3_K_M | 187 tok/s | — | 1 |
| Qwen3-1.7B | llama.cpp | IQ2_M | 311.6 tok/s | — | 1 |
| Qwen3-1.7B | llama.cpp | Q5_K_S | 197.9 tok/s | — | 1 |
| Qwen3-1.7B | vLLM | Q4_K_M | 233.4 tok/s | — | 1 |
| Qwen3-1.7B | vLLM | Q5_K_S | 229.8 tok/s | — | 1 |
| Qwen3-4B-Thinking-2507 | llama.cpp | Q3_K_L | 181 tok/s | — | 1 |
| Qwen2.5-0.5B-Instruct | vLLM | Q5_K_S | 376.6 tok/s | — | 1 |
| Qwen2.5-0.5B-Instruct | vLLM | Q8_0 | 284.1 tok/s | — | 1 |
| Qwen2.5-Omni-3B | Ollama | Q5_K_M | 149.1 tok/s | — | 1 |
| Qwen3-Embedding-0.6B | llama.cpp | IQ1_S | 286.7 tok/s | — | 1 |
| Qwen3-Embedding-0.6B | llama.cpp | IQ2_M | 301.8 tok/s | — | 1 |
| Qwen3-Embedding-0.6B | Ollama | Q4_K_S | 342.8 tok/s | — | 1 |
| gemma-4-31B-it | Ollama | Q3_K_M | 62.3 tok/s | — | 1 |
| gemma-4-31B-it | Strata | IQ3_XS | 111.4 tok/s | — | 1 |
| gemma-4-26B-A4B-it | llama.cpp | Q8_0 | 55.3 tok/s | — | 1 |
| gemma-4-26B-A4B-it | Strata | Q6_K | 98.9 tok/s | — | 1 |
| gemma-4-12B-it | llama.cpp | Q4_K_M | 100.4 tok/s | — | 1 |
| gemma-4-12B-it | llama.cpp | Q8_0 | 66.9 tok/s | — | 1 |
| gemma-4-12B-it | Ollama | Q3_K_M | 110.4 tok/s | — | 1 |
| gemma-4-12B-it | Ollama | Q3_K_S | 111.3 tok/s | — | 1 |
| DeepSeek-R1-Distill-Qwen-14B | llama.cpp | Q2_K | 127.4 tok/s | — | 1 |
| GLM-5.3 | llama.cpp | IQ2_XS | 0.9 tok/s | — | 1 |
| GLM-5.3 | llama.cpp | IQ3_S | 0.8 tok/s | — | 1 |
| DeepSeek-V4.1-Flash | llama.cpp | IQ2_XS | 0.9 tok/s | — | 1 |
| DeepSeek-V4.1-Flash | llama.cpp | IQ3_S | 0.8 tok/s | — | 1 |
| Qwen3.8-27B | llama.cpp | IQ2_XS | 26.3 tok/s | — | 1 |
| Qwen3.8-27B | llama.cpp | Q5_K_M | 12 tok/s | — | 1 |
| Qwen3.8-27B | llama.cpp | Q4_K_M | 14.6 tok/s | — | 1 |