phi-3-mini 3.8B
phi-3-mini 3.8B is an open-weights language model from Microsoft, released April 23, 2024. It has 3.8B parameters (dense). On the Epoch Capabilities Index it scores 110.6, rank 203 of 256 rated models.
Check it against your own PC →
Can you run phi-3-mini 3.8B locally?
The smallest common GPU it runs on is 6 GB of VRAM (with 32 GB of RAM): Q5_K_M, fully on GPU, roughly 41 tokens per second. Estimates assume 32 GB of DDR5 RAM, a 16K context and llama.cpp-style GGUF quantization. Enter your exact hardware in the live app.
| GPU memory | Verdict | Best quant | VRAM | RAM | tok/s |
|---|---|---|---|---|---|
| 6 GB VRAM | Runs on GPU | Q5_K_M (excellent) | 5.1 GB | – | ~41 |
| 8 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 5.5 GB | – | ~50 |
| 12 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 5.5 GB | – | ~65 |
| 16 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 5.5 GB | – | ~88 |
| 24 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 5.5 GB | – | ~150 |
| 32 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 5.5 GB | – | ~260 |
Benchmark results
| Benchmark (Epoch AI) | Score |
|---|---|
| MMLU | 58.4% |
Links
Models with a similar score
phi-3-medium 14B · Claude Instant · PaLM 2-L · Qwen2.5-Coder-14B · Mistral NeMo · Mixtral 8x7B · GPT-3.5 Turbo (Jan 2024) · GPT-3.5 Turbo (Nov 2023)
More from Microsoft
MAI-Cyber-1.1-Flash · MAI-Base-2 · MAI-Cyber-1-Flash · MAI-Code-1.1-Flash · Mage VL · MAI-Thinking-1 · MAI-Code-1-Flash · Phi-4