ChatCodingAI CodingReasoning3.82B
Phi-4 Mini
Phi · MIT
Microsoft's tiny powerhouse. Punches well above its weight at 3.8B parameters — competitive with many 7B models on reasoning and coding benchmarks. Extremely fast inference. Ideal as a draft model or for resource-constrained setups.
- Parameters
- 3.82B
- Context
- 16,384 tokens
- Released
- 2025-02-27
- Engines
- llama.cpp, ollama, vLLM
- Builder Tools
- Cursor, Continue, Ollama, LM Studio
Phi-4 Mini (3.82B) requires 3.3 GB VRAM at recommended quality (Q6_K). At efficient quality (Q4_K_M), it fits in 2.4 GB VRAM, making it compatible with the NVIDIA GeForce RTX 3060 12GB. On NVIDIA GeForce RTX 5090, expect approximately 185 tok/s at Q8_0. For the best experience, Starter AI Desktop ($582) is recommended.
— OwnRig methodology, data updated 2026-03-15
VRAM Requirements
| Quality | Quantization | VRAM | File Size |
|---|---|---|---|
| full | Q8_0 | 4.3 GB | 3.8 GB |
| recommended | Q6_K | 3.3 GB | 2.9 GB |
| recommended | Q5_K_M | 2.8 GB | 2.5 GB |
| efficient | Q4_K_M | 2.4 GB | 2.1 GB |
| compressed | Q3_K_M | 2 GB | 1.7 GB |
Context Length Impact
KV cache VRAM at Q6_K quality. Longer context = more memory.
| Context | KV Cache | Total VRAM |
|---|---|---|
| 2K | 102 MB | 3.4 GB |
| 4K | 102 MB | 3.4 GB |
| 8K | 307 MB | 3.6 GB |
| 16K | 512 MB | 3.8 GB |
Compatible GPUs
Performance data for Phi-4 Mini across different hardware.
| Device | Quantization | Speed | Rating | Fits in VRAM |
|---|---|---|---|---|
| NVIDIA GeForce RTX 3060 12GB | Q8_0 | 80 tok/s | Excellent | ✓ |
| NVIDIA GeForce RTX 4060 Ti 16GB | Q8_0 | 68 tok/s | Excellent | ✓ |
| NVIDIA GeForce RTX 4070 Ti Super | Q8_0 | 120 tok/s | Excellent | ✓ |
| NVIDIA GeForce RTX 4070 Super | Q8_0 | 100 tok/s | Excellent | ✓ |
| NVIDIA GeForce RTX 4080 Super | Q8_0 | 130 tok/s | Excellent | ✓ |
| NVIDIA GeForce RTX 4090 | Q8_0 | 160 tok/s | Excellent | ✓ |
| NVIDIA GeForce RTX 3090 | Q8_0 | 140 tok/s | Excellent | ✓ |
| NVIDIA GeForce RTX 5080 | Q8_0 | 150 tok/s | Excellent | ✓ |
| NVIDIA GeForce RTX 5090 | Q8_0 | 185 tok/s | Excellent | ✓ |
| Apple M4 Pro (24GB Unified) | Q8_0 | 55 tok/s | Excellent | ✓ |
| Apple M4 Pro (48GB) | Q8_0 | 55 tok/s | Excellent | ✓ |
| Apple M4 Max (36GB Unified) | Q8_0 | 90 tok/s | Excellent | ✓ |
| Apple M4 Max (64GB Unified) | Q8_0 | 90 tok/s | Excellent | ✓ |
| Apple M4 Max (128GB Unified) | Q8_0 | 90 tok/s | Excellent | ✓ |
| NVIDIA GeForce RTX 4060 8GB | Q5_K_M | 55 tok/s | Excellent | ✓ |
| NVIDIA GeForce RTX 4070 Ti 12GB | Q8_0 | 82 tok/s | Excellent | ✓ |
| NVIDIA GeForce RTX 3080 10GB | Q8_0 | 120 tok/s | Excellent | ✓ |
| Apple M3 Pro (18GB Unified) | Q8_0 | 30 tok/s | Good | ✓ |
Builder Context
Phi-4 Mini is commonly used with Cursor, Continue, Ollama, LM Studio. For an AI coding workflow, pair it with an embedding model like nomic-embed-text for local RAG.
Recommended Builds
Complete PC builds that can run Phi-4 Mini.
Frequently Asked Questions
- How much VRAM does Phi-4 Mini need?
- Phi-4 Mini requires 3.3 GB VRAM at recommended quality (Q6_K). At lower quality settings, it can fit in as little as 2 GB.
- What is the best GPU for Phi-4 Mini?
- The NVIDIA GeForce RTX 5090 delivers the best performance for Phi-4 Mini, achieving 185 tok/s at Q8_0 with an excellent rating.
- Can I run Phi-4 Mini on an RTX 4060 Ti?
- Yes. On the NVIDIA GeForce RTX 4060 Ti 16GB, Phi-4 Mini runs at 68 tok/s (Q8_0, excellent).
- What quantization should I use for Phi-4 Mini?
- For the best quality, use Q6_K (3.3 GB VRAM). If your GPU has limited VRAM, Q3_K_M (2 GB) is the most efficient option with acceptable quality.
- Is Phi-4 Mini good for coding?
- Yes. Phi-4 Mini is used with Cursor, Continue, Ollama, LM Studio for local AI coding. For the best coding experience, pair it with an embedding model for local RAG.
Data confidence: estimated. Last updated: 2026-03-15. Source