Deploy Phi-4 mini as an OpenAI-compatible API
ModelDock launches Hugging Face models behind an OpenAI-compatible API. Phi-4 mini (compact model tuned for reasoning and code) is a 3.8B-parameter model from Microsoft, released under the mit license. Paste its link in the dashboard and Pilot deploys it as a free 15-minute test, then you call it with any OpenAI SDK.
Phi-4 mini at a glance
| Hugging Face repo | microsoft/Phi-4-mini-instruct |
|---|---|
| Made by | Microsoft |
| Parameters | 3.8B |
| License | mit |
| Context length (config.json) | 131,072 tokens |
| Hugging Face downloads | 343K (as of 2026-10-25) |
| Typical GPU | 16 GB: RTX A4000 / A4500 / 4000 Ada |
| Output price per 1M tokens | $1.21 |
| Input price per 1M tokens | $0.30 |
| Start fee | $0.021 |
// facts from the Hugging Face model page. GPU tier and prices are Pilot’s typical pick for 16-bit weights using the pricing formula, not a guarantee; Pilot makes the final choice at launch, and paid launches are opening soon.
Launch it in three steps
- Sign in with your litepod account and paste https://huggingface.co/microsoft/Phi-4-mini-instruct in the dashboard.
- Pilot checks the model, picks a GPU and deploys it. Copy the model id and an API key when it is ready.
- Call it with the code below. The free test lasts 15 minutes or 25 requests.
from openai import OpenAI
client = OpenAI(base_url="https://api.lab.litepod.io/v1", api_key="sk_modeldock_...")
reply = client.chat.completions.create(
model="modeldock/<your-model-id>", # shown in the dashboard after launching microsoft/Phi-4-mini-instruct
messages=[{"role": "user", "content": "Hello!"}],
)
print(reply.choices[0].message.content)More in the quickstart, OpenAI compatibility and limits. See pricing for every GPU tier.
Similar models
- Qwen3 4B
Alibaba Qwen · 4B · apache-2.0
- SmolLM3 3B
Hugging Face · 3.1B · apache-2.0
- Qwen3 1.7B
Alibaba Qwen · 2B · apache-2.0
All models · More from litepod: the free LLM API with free open-source models