Phi-4: price on every cloud
Open weights · 14 billion parameters · 16,384-token context · licence mit
Per token, by cloud
| Cloud | Input, per 1M tokens | Output, per 1M tokens | Cheapest region |
|---|---|---|---|
| Azure | $0.075 | $0.30 | centralus · 13 regions |
| AWS | — | — | not sold per token |
| OCI | — | — | not sold per token |
| Core42 | — | — | not sold per token |
| Alibaba Cloud | — | — | not sold per token |
| Google Cloud | — | — | not sold per token |
Run it yourself
It needs 36.6 GB of GPU memory at bf16, the precision it ships in.
| Cloud | Cheapest instance that fits | GPUs | An hour | A month, around the clock |
|---|---|---|---|---|
| AWS | g6e.xlarge | 1 × L40S 48 GB | $1.86 | $1,359 |
| Alibaba Cloud | ecs.gn8is.2xlarge | 1 × L20 48 GB | $2.26 | $1,648 |
| Azure | Standard_NC72ds_xl_RTXPRO6000BSE_v6 | 1 × RTX PRO 6000 48 GB | $2.83 | $2,066 |
| Google Cloud | a2-highgpu-1g | 1 × A100 40 GB | $3.67 | $2,682 |
At the cheapest per-token price, running it on AWS g6e.xlarge pays off above roughly 10.4 billion tokens a month.
More from Microsoft
- Phi-3.5-vision-instruct
- phi-2
- Phi-3.5-mini-instruct
- Phi-4-mini-instruct
- Phi-3-mini-4k-instruct
- Phi-4-multimodal-instruct
Similar size, open weights
Inside the UAE: no cloud sells it in-country.
List prices collected 5 October 2026, refreshed every three hours. No negotiated rates or taxes.