
Grok 3 API 已經發布,它的“scary‑smart”推理、即時網路功能,以及在編碼和 STEM 基準測試中的頂級效能,已經撼動了人工智慧世界。無論您是開發人員、研究人員還是人工智慧愛好者,現在都是訪問 Grok 3 API 並探索其功能的最佳時機。在本教程中,您將學習如何註冊使用 Grok 3 API、安全地驗證您的憑據以及使用 Python 連線 Grok 3 API。我們還將介紹一些實際用例,從將 Grok 3 API 整合到工作流程,到使用 Grok 3 API 進行資料檢索和高階推理任務。
Grok 3的主要功能
Grok 3 引入了幾項突破性功能,這些功能增強了 Grok 3 在各個領域的效能和適用性:
- 高階推理模式:Grok 3 提供了專門的推理模式,包括用於逐步解決問題的 “思考 ”模式和用於處理複雜任務的 “大大腦 ”模式。這些模式可使模型更深入地處理資訊,並提供準確的響應。
- 深度搜尋功能:該模型整合了 DeepSearch,這是一種人工智慧代理,可即時掃描網際網路和 X(前 Twitter),生成有關特定主題的綜合報告。該功能可確保 Grok 3 的回覆與時俱進、資訊靈通。
- 增強的效能基準:Grok 3 在 2025 年美國數學邀請考試(AIME)中取得了 93.3% 的高分,並在聊天機器人競技場中取得了 1402 的 Elo 分數,顯示了其在 STEM 領域的優勢。
- 即時資料整合:與靜態的人工智慧模型不同,Grok 3 整合了來自網路和 X 帖子的即時資料,確保其回覆是最新的、相關的。
- 多模式功能:該模型可以處理和生成文字、影像和程式碼,從而將其適用範圍擴充套件到各個領域。
這些功能使 Grok 3 成為人工智慧語言模型的一大進步。它提供了增強的推理能力、即時資料處理能力和廣泛的功能,可滿足不同使用者的需求。
價格和模型規格
xAI 推出了 Grok-3 模型系列的多個變體。每種模型都針對不同的效能和成本效益水平進行了微調。這些模型可滿足開發人員、研究人員和組織機構的計算和推理需求。

Source: X.com
在呼叫API之前使用xAI的成本計算器
在您開始呼叫多個 API 之前,xAI 為其每種模型(包括 grok-3-beta、grok-3-fast-beta 和迷你變體)提供了一個超級方便的成本計算器。
您可以直接訪問 https://x.ai/api(向下滾動即可)。

您可以自定義
- 文字輸入 token
- 影像輸入 token
- 輸出 token
該工具可為您提供即時成本估算,以便您有效規劃使用量,避免意外計費。例如,在 grok-3-beta 上使用 50,000 個文字輸入 token 和 500 個輸出 token,只需花費 0.16 美元(如前面的截圖所示)。
智慧開發提示:經常使用此計算器來最佳化您的 API 策略,尤其是在擴充套件或處理大型有效載荷時。
如何訪問API?
1. 訪問 grok.com,使用您的賬戶憑據登入。

2. 點選右上角的個人資料頭像,選擇“Settings”,然後選擇“Manage”。您將被重定向到 xAI 賬戶頁面。


3. 在 xAI 賬戶頁面,導航至 API Console。

4. 在 API 控制檯的左側邊欄,點選金鑰圖示,檢視並複製您的 Grok API 金鑰。

瞧,這就是你的 API 金鑰!請務必妥善保管。
Grok 3的實現
基本實現
讓我們嘗試使用 Grok xAI 文件中提供的程式碼片段來檢查我們的 Grok 3 模型是否能夠響應。
os.environ['GROK_API_KEY'] = "xai-..." # your own api key
from IPython.display import Markdown
from openai import OpenAI
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
completion = client.chat.completions.create(
model="grok-3-mini-beta",
{"role": "user", "content": "What is the meaning of life?"}
Markdown(completion.choices[0].message.content)
!pip install openai
import os
os.environ['GROK_API_KEY'] = "xai-..." # your own api key
from IPython.display import Markdown
from openai import OpenAI
client = OpenAI(
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
)
completion = client.chat.completions.create(
model="grok-3-mini-beta",
messages=[
{"role": "user", "content": "What is the meaning of life?"}
]
)
Markdown(completion.choices[0].message.content)
!pip install openai
import os
os.environ['GROK_API_KEY'] = "xai-..." # your own api key
from IPython.display import Markdown
from openai import OpenAI
client = OpenAI(
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
)
completion = client.chat.completions.create(
model="grok-3-mini-beta",
messages=[
{"role": "user", "content": "What is the meaning of life?"}
]
)
Markdown(completion.choices[0].message.content)
輸出:

現在,讓我們測試一下 Grok 3 模型: – 生成程式碼
- 程式碼生成
- 推理能力
- 複雜用例
- 科學研究理解
您可以隨意更換其他 Grok 3 變體(例如,grok-3-beta、grok-3-fast-beta)或製作您的提示。然後,在筆記本或指令碼中直接比較輸出結果。
在下面的實現中,我們將利用
程式碼生成
1. 讓我們使用 Grok 3 模型生成程式碼,將華氏度轉換為攝氏度,反之亦然。
Write a Python function that converts a temperature from Fahrenheit to Celsius and vice versa.
The function should take an input, determine the type (Fahrenheit or Celsius), and return the converted temperature.
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
response = client.chat.completions.create(
model="grok-3-mini-beta",
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
prompt = """
Write a Python function that converts a temperature from Fahrenheit to Celsius and vice versa.
The function should take an input, determine the type (Fahrenheit or Celsius), and return the converted temperature.
"""
client = OpenAI(
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
)
response = client.chat.completions.create(
model="grok-3-mini-beta",
messages=[
{
"role": "user",
"content": prompt
}
]
)
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
prompt = """
Write a Python function that converts a temperature from Fahrenheit to Celsius and vice versa.
The function should take an input, determine the type (Fahrenheit or Celsius), and return the converted temperature.
"""
client = OpenAI(
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
)
response = client.chat.completions.create(
model="grok-3-mini-beta",
messages=[
{
"role": "user",
"content": prompt
}
]
)
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
輸出:

2. 讓我們用 Grok 3 模型生成一個 HTML 頁面,頁面上有一個按鈕,點選後頁面上就會灑滿紙屑。
prompt = """Create an HTML page with a button that explodes confetti when you click it.
You can use CSS & JS as well."""
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
response = client.chat.completions.create(
model="grok-3-mini-beta",
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
prompt = """Create an HTML page with a button that explodes confetti when you click it.
You can use CSS & JS as well."""
client = OpenAI(
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
)
response = client.chat.completions.create(
model="grok-3-mini-beta",
messages=[
{
"role": "user",
"content": prompt
}
]
)
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
prompt = """Create an HTML page with a button that explodes confetti when you click it.
You can use CSS & JS as well."""
client = OpenAI(
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
)
response = client.chat.completions.create(
model="grok-3-mini-beta",
messages=[
{
"role": "user",
"content": prompt
}
]
)
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
輸出:

為了測試這段程式碼的功能,我將使用 CodePen 來測試 HTML 程式碼。
這就是它的外觀:
推理能力
1. 讓我們用這道通用能力推理題來測試一下 Grok 3 模型
prompt = """Anu is a girl. She has three brothers. Each of her brothers has the same two sisters.
How many sisters does Anu have?"""
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
response = client.chat.completions.create(
model="grok-3-mini-beta",
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
prompt = """Anu is a girl. She has three brothers. Each of her brothers has the same two sisters.
How many sisters does Anu have?"""
client = OpenAI(
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
)
response = client.chat.completions.create(
model="grok-3-mini-beta",
messages=[
{
"role": "user",
"content": prompt
}
]
)
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
prompt = """Anu is a girl. She has three brothers. Each of her brothers has the same two sisters.
How many sisters does Anu have?"""
client = OpenAI(
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
)
response = client.chat.completions.create(
model="grok-3-mini-beta",
messages=[
{
"role": "user",
"content": prompt
}
]
)
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
輸出:

已驗證答案正確。
2. 讓我們來測試一下 Grok 3 的模式識別能力,提供這個基於日期的模式問題
prompt = """January = 1017, February = 628, March = 1335, April = 145, May = 1353, June = 1064,
July = 1074, August = 186, September = ? Think carefully before answering also show the steps"""
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
response = client.chat.completions.create(
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
prompt = """January = 1017, February = 628, March = 1335, April = 145, May = 1353, June = 1064,
July = 1074, August = 186, September = ? Think carefully before answering also show the steps"""
client = OpenAI(
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
)
response = client.chat.completions.create(
model="grok-3-beta",
messages=[
{
"role": "user",
"content": prompt
}
]
)
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
prompt = """January = 1017, February = 628, March = 1335, April = 145, May = 1353, June = 1064,
July = 1074, August = 186, September = ? Think carefully before answering also show the steps"""
client = OpenAI(
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
)
response = client.chat.completions.create(
model="grok-3-beta",
messages=[
{
"role": "user",
"content": prompt
}
]
)
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
輸出:

GROK-BETA 和 GROK-BETA-MINI 都失敗了
答案應為 1999。
3. 讓我們用這個問題來測試一下 Grok 3 模型的簡單數學推理能力吧
prompt = """I have two apples, then I buy two more. I bake a pie with two of the apples.
After eating half of the pie, how many apples do I have left?"""
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
response = client.chat.completions.create(
model="grok-3-mini-beta",
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
/cod
prompt = """I have two apples, then I buy two more. I bake a pie with two of the apples.
After eating half of the pie, how many apples do I have left?"""
client = OpenAI(
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
)
response = client.chat.completions.create(
model="grok-3-mini-beta",
messages=[
{
"role": "user",
"content": prompt
}
]
)
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
/cod
prompt = """I have two apples, then I buy two more. I bake a pie with two of the apples.
After eating half of the pie, how many apples do I have left?"""
client = OpenAI(
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
)
response = client.chat.completions.create(
model="grok-3-mini-beta",
messages=[
{
"role": "user",
"content": prompt
}
]
)
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
輸出:

已驗證答案正確。
複雜案例
讓我們要求 Grok 3 模型為我們提供一個完整的專案結構計劃和相應的程式碼片段,這些程式碼片段必須相應地新增到相應的檔案中。
I want to develop an inventory management system that tracks products, quantities,
and locations. It should notify the user when stock is low. Create a plan for the
directory structure and provide code snippets for the key components.
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
response = client.chat.completions.create(
model="grok-3-mini-beta",
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
prompt = """
I want to develop an inventory management system that tracks products, quantities,
and locations. It should notify the user when stock is low. Create a plan for the
directory structure and provide code snippets for the key components.
"""
client = OpenAI(
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
)
response = client.chat.completions.create(
model="grok-3-mini-beta",
messages=[
{
"role": "user",
"content": prompt
}
]
)
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
prompt = """
I want to develop an inventory management system that tracks products, quantities,
and locations. It should notify the user when stock is low. Create a plan for the
directory structure and provide code snippets for the key components.
"""
client = OpenAI(
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
)
response = client.chat.completions.create(
model="grok-3-mini-beta",
messages=[
{
"role": "user",
"content": prompt
}
]
)
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
輸出:

科學研究理解
在此,我們將測試 Grok 3 模型基於領域的理解能力,以及它是如何理解問題和闡述某些科學研究主題的。
Explain how CRISPR technology can be used to treat genetic disorders. What are the
main challenges, and what future advancements might be necessary to make it widely
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
response = client.chat.completions.create(
model="grok-3-mini-beta",
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
prompt = """
Explain how CRISPR technology can be used to treat genetic disorders. What are the
main challenges, and what future advancements might be necessary to make it widely
available?
"""
client = OpenAI(
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
)
response = client.chat.completions.create(
model="grok-3-mini-beta",
messages=[
{
"role": "user",
"content": prompt
}
]
)
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
prompt = """
Explain how CRISPR technology can be used to treat genetic disorders. What are the
main challenges, and what future advancements might be necessary to make it widely
available?
"""
client = OpenAI(
api_key=os.getenv("GROK_API_KEY"),
base_url="https://api.x.ai/v1",
)
response = client.chat.completions.create(
model="grok-3-mini-beta",
messages=[
{
"role": "user",
"content": prompt
}
]
)
# Output the generated Python code
print(response.choices[0].message.content)
Markdown(response.choices[0].message.content)
輸出:

我的觀點
編碼能力
我發現 Grok 3 的程式碼生成準確度令人印象深刻。在典型的演算法任務上,它的表現優於 gpt-4o、Deepseek-R1 等較輕的模型,不過我建議在下結論之前,先在更復雜的場景中對它進行壓力測試。

推理能力
Grok 3 具有很強的數學推理能力,能夠可靠地解決多步驟問題。不過,我注意到在基於模式的推理方面出現了一些失誤,如識別不明顯的類比或隱藏序列。
複雜用例和科學理解
當被要求概述專案結構時,Grok 3 提供了一份條理清晰的計劃,並附有模板程式碼片段。它對 CRISPR 應用的瞭解給我留下了深刻印象,因為它提供了詳細的概述,顯示了對科學的深刻理解。
Grok 3 擅長闡述,並能清晰、深入地給出答案。它在 LLM 推理方面邁出了一大步,但我們還是應該一如既往地根據您的基準來驗證它的輸出結果。
小結
Grok 3 是大型語言模型發展史上的一個重要里程碑。從其“scary‑smart”推理模式和即時 DeepSearch 功能,到多模態支援和業界領先的基準測試,xAI 推出的工具包不僅功能強大,而且用途廣泛。透過本指南,您將學會如何
- 保護您的 Grok 3 API 金鑰並確保其安全
- 使用 xAI 的內建計算器預先估算成本
- 在幾分鐘內完成基本聊天
- 在程式碼生成、邏輯謎題、複雜工作流和科學查詢中試用 Grok 3。
- 評估其優勢並確定需要進一步壓力測試的領域
無論您是開發下一個偉大聊天機器人的開發人員,還是探索人工智慧驅動分析的研究人員,抑或僅僅是對LLM 尖端技術充滿好奇的愛好者,Grok 3 都能將深度、速度和現實世界的相關性完美結合。
當您將 Grok 3 整合到您的專案中時,請記得根據您的領域基準驗證其輸出結果,最佳化令牌的使用,並與社羣分享您的發現。編碼快樂 🙂
評論留言