Written by Jakub Rusinowski · Last updated August 20, 2024
Model library → Phi 3.5 Family → Phi 3.5 Mini
Beats Llama 3 8B in some benchmarks while running on a phone.
Phi 3.5 Mini needs about 3 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache. On Apple Silicon that figure comes out of unified memory.
| Parameters | 3.8 Billion |
| Context window | 128,000 |
| Architecture | Dense |
| Provider | Microsoft |
| Licence | MIT |
| Specified at | Q4_K_M |
| System RAM | 8 GB |
| Record updated | 2024-08-20 |
MIT — commercial use permitted. Commercial use permitted. No usage restrictions beyond attribution.
Modelled on a reference NVIDIA RTX 4090 (24 GB), at 8K context. Speed figures are ESTIMATES from the memory-bandwidth roofline described on the methodology page, not benchmarks we ran — rows marked measured come from published or reader-submitted runs. VRAM here includes the KV cache, so it reads higher than the headline figure above, which does not.
| Quant | Weights | VRAM needed | Est. speed | Fit on 24 GB |
|---|---|---|---|---|
| Q2_K | 1.2 GB | 2.9 GB | ~226 tok/s (est.) | Fits comfortably |
| Q3_K_M | 1.6 GB | 3.2 GB | ~203 tok/s (est.) | Fits comfortably |
| Q4_K_M | 2.3 GB | 3.9 GB | ~172 tok/s (est.) | Fits comfortably |
| Q5_K_M | 2.7 GB | 4.3 GB | ~158 tok/s (est.) | Fits comfortably |
| Q6_K | 3.1 GB | 4.7 GB | ~145 tok/s (est.) | Fits comfortably |
| Q8_0 | 4.0 GB | 5.6 GB | ~123 tok/s (est.) | Fits comfortably |
| F16 | 7.6 GB | 9.2 GB | ~78 tok/s (est.) | Fits comfortably |
Want the memory numbers alone, at every quantization level and your own context length? Use the Phi 3.5 Mini VRAM calculator.
or compare on Vast.ai from $0.35/hr (typical low · varies)
As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.
The cheapest catalogued GPU that runs Phi 3.5 Mini is the Intel Arc B570 (10 GB).
Install Ollama, then run:
ollama run phi3.5Weights on Hugging Face: microsoft/Phi-3.5-mini-instruct.
Best for: mobile, fast chat.
← All Phi 3.5 Family models | VRAM calculator | Check your own hardware