First Call¶

The smallest possible round-trip to an LLM: send one message, print the reply. Shown for all three providers we cover — all of them use the same openai client.

Prerequisites¶

A key from at least one provider. If you don't have one, pick from the Get an API Key page.
The key stored in your environment. We'll assume a .env file at the project root with whichever you have:
```
OPENAI_API_KEY=sk-...
DEEPSEEK_API_KEY=sk-...
DASHSCOPE_API_KEY=sk-...
```
The llm-tutorial conda env activated:
```
conda activate llm-tutorial
```

The openai SDK is already installed as a project dependency. DeepSeek and Qwen reuse the openai SDK — only base_url and model change.

Hello world¶

SJTUOpenAIDeepSeekQwen

SJTU's OpenAI-compatible gateway at models.sjtu.edu.cn exposes deepseek-chat and other models (campus-network / VPN only). Same code shape as DeepSeek — just swap the key env var and the base URL.

from openai import OpenAI
client = OpenAI(
    api_key=os.environ["SJTU_API_KEY"],
    base_url="https://models.sjtu.edu.cn/api/v1",
)
model = "deepseek-chat"

first_call_openai.py

import os
from dotenv import load_dotenv
from openai import OpenAI

load_dotenv()
client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

resp = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[
        {"role": "user", "content": "Say hi in one short sentence."},
    ],
)

print(resp.choices[0].message.content)

Expected output:

Hi there! How can I help you today?

DeepSeek exposes an OpenAI-compatible endpoint, so the openai client works as-is — just override base_url and pick a DeepSeek model.

first_call_deepseek.py

import os
from dotenv import load_dotenv
from openai import OpenAI

load_dotenv()
client = OpenAI(
    api_key=os.environ["DEEPSEEK_API_KEY"],
    base_url="https://api.deepseek.com",
)

resp = client.chat.completions.create(
    model="deepseek-chat",
    messages=[
        {"role": "user", "content": "Say hi in one short sentence."},
    ],
)

print(resp.choices[0].message.content)

Expected output:

Hello! Hope you're doing well.

Use model="deepseek-reasoner" for the reasoning model; the response will additionally carry resp.choices[0].message.reasoning_content.

Qwen on Alibaba's DashScope also exposes an OpenAI-compatible endpoint.

first_call_qwen.py

import os
from dotenv import load_dotenv
from openai import OpenAI

load_dotenv()
client = OpenAI(
    api_key=os.environ["DASHSCOPE_API_KEY"],
    base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
)

resp = client.chat.completions.create(
    model="qwen-plus",
    messages=[
        {"role": "user", "content": "Say hi in one short sentence."},
    ],
)

print(resp.choices[0].message.content)

Expected output:

Hi! Nice to meet you.

Response shape¶

Most fields you'll touch:

resp.choices[0].message.content    # the assistant's text
resp.choices[0].finish_reason      # "stop", "length", "tool_calls", ...
resp.model                         # resolved model ID
resp.usage.prompt_tokens           # input tokens billed
resp.usage.completion_tokens       # output tokens billed
resp.usage.total_tokens

Adding a system prompt¶

Prepend a message with role: "system":

resp = client.chat.completions.create(
    model="gpt-4o-mini",  # or "deepseek-chat", "qwen-plus"
    messages=[
        {"role": "system", "content": "You are a concise assistant. Answer in one sentence."},
        {"role": "user", "content": "What is a PID controller?"},
    ],
)

Troubleshooting¶

AuthenticationError / 401: your key is wrong, revoked, or scoped to a different workspace / region.
OpenAI insufficient_quota: the key has no credits; top up at Billing.
ConnectionError: check your network / proxy; some provider endpoints are region-blocked.

Next¶

Unified Client — wrap all three providers behind one function so you can swap with a single line.
Streaming — return tokens as they're generated.
Tool Use — let the model call your Python functions.