Phi Models: Small AI That Thinks Big
Microsoft's Phi family proves that size is not everything. The local lineup spans four variants: Phi-3 Mini and Phi-4 Mini at 3.8B, plus Phi-3 Medium and Phi-4 at 14B. Phi-4 Mini loads in about 3.2GB and runs on any 8GB Mac, making it one of the best low-RAM picks we track. At the top, Phi-4 14B needs a 22GB machine and delivers reasoning quality near much bigger models. If your RAM is limited, Phi gives you the most intelligence per gigabyte.
- developer
- Microsoft
- local models
- 6
- size range
- 3.8B-14B
- RAM range
- 6-24 GB
All Phi models
6 local models| Model | Size | Quant | VRAM | Min RAM | Best For | Quality | Ollama |
|---|---|---|---|---|---|---|---|
| Phi-3 Mini 3.8B | 3.8B | Q4_K_M | 3.2 GB | 6 GB | Coding, Chat | 65 | |
| Phi-4 Mini 3.8B | 3.8B | Q4_K_M | 3.2 GB | 6 GB | Coding, Chat | 73 | |
| Phi-4 Mini 3.8B (Q8) | 3.8B | Q8_0 | 3.8 GB | 8 GB | Coding, Chat | 75 | |
| Phi-3 Medium 14B | 14B | Q4_K_M | 11 GB | 16 GB | Coding, Quality | 75 | |
| Phi-4 14B | 14B | Q4_K_M | 11.5 GB | 24 GB | Coding, Quality | 86 | |
| Phi-4 14B (Q8) | 14B | Q8_0 | 14.5 GB | 24 GB | Coding, Quality | 88 |
Key features
- Best quality-per-gigabyte at small sizes
- Phi-4 Mini 3.8B loads in just 3.2GB (7GB min RAM)
- Phi-4 14B delivers high reasoning quality from an 11.5GB load
- Strong coding ability across the whole lineup
- Runs on 8GB MacBook Airs and other low-RAM devices
- Simple Ollama setup with phi4-mini and phi4 tags
Device compatibility
Which Phi models can run on each device class, based on minimum RAM requirements.
| Model | iPhone | Air | Pro | Studio | Mini | |
|---|---|---|---|---|---|---|
| Phi-3 Mini 3.8B (3.8B) | Possible | Possible | Excellent | Excellent | Excellent | Possible |
| Phi-4 Mini 3.8B (3.8B) | Possible | Possible | Excellent | Excellent | Excellent | Possible |
| Phi-4 Mini 3.8B (Q8) (3.8B) | Possible | Possible | Excellent | Excellent | Excellent | Possible |
| Phi-3 Medium 14B (14B) | No | Possible | Possible | Excellent | Possible | Possible |
| Phi-4 14B (14B) | No | Possible | Possible | Possible | Possible | Possible |
| Phi-4 14B (Q8) (14B) | No | Possible | Possible | Possible | Possible | Possible |
RAM requirements
Largest dense Q4 model that fits each memory tier
Get told when a better model fits your MacBook Air M5 16 GB
Get one email when a new open-weight model is a better fit for this machine.
For: MacBook Air M5 16 GB
You'll also get the weekly ModelFit email. By signing up you agree to our Privacy Policy. Unsubscribe from either email anytime, in one click.
Config reports
No speed test run for this family yet. Run the speed test: npx @modelfit/cli@latest bench
Share your config
Frequently asked questions
What is the difference between Phi-4 and Phi-4 Mini?
Phi-4 Mini is the 3.8B model: a 3.2GB load and a 7GB RAM minimum, ideal for 8GB Macs. Phi-4 is the 14B model: an 11.5GB load and a 22GB minimum, so it wants a 24GB or 32GB machine.
What RAM does each Phi model need?
Phi-3 Mini and Phi-4 Mini both need 7GB minimum with loads around 3.2GB. Phi-3 Medium 14B needs 20GB and Phi-4 14B needs 22GB. All figures assume the standard Q4 quantization used by Ollama.
What Ollama command runs Phi?
Run `ollama run phi4-mini:3.8b` for the small model or `ollama run phi4:14b-q4_K_M` for the 14B. The previous generation is still available via `ollama run phi3:mini` and `ollama run phi3:medium`.
Is Phi good for coding?
Yes. Phi-4 Mini is a fast, surprisingly capable coding assistant on low-RAM Macs. Phi-4 14B offers high coding quality with slower throughput, making it a solid pick for 24GB machines.
