1. What is DeepSeek V4?

DeepSeek V4 (model name: deepseek-chat) is the fourth-generation flagship model from DeepSeek AI, a Chinese research lab that shocked the AI world in early 2025 by releasing GPT-4-class models at a fraction of the training cost.

V4 continues that tradition: it delivers performance comparable to GPT-4o and Claude Sonnet 4 on most developer benchmarks, while pricing starts at $0.27 per million input tokens — roughly 10x cheaper than GPT-4o.

Key facts:

• Architecture: Mixture-of-Experts (MoE) — activates only a fraction of parameters per forward pass
• Context window: 128K tokens
• API compatibility: OpenAI-compatible (drop-in replacement)
• Modalities: Text only (no native vision in deepseek-chat)
• Availability: Cloud API + open weights for self-hosting

2. Benchmarks: How Good Is It?

Takeaway: DeepSeek V4 matches or beats GPT-4o on coding and math tasks. Claude Sonnet leads on complex reasoning and agentic tasks. Gemini excels on long-context. For pure price-performance, DeepSeek wins decisively.

3. Pricing — The Real Story

At $0.27/1M input tokens, a typical agent workflow processing 500K tokens per day costs $4/month — vs $37.50/month with GPT-4o. The savings compound fast at scale.

4. API Quickstart (5 Minutes)

Get your API key

Sign up at platform.deepseek.com → API Keys → Create Key. New accounts get free credits.

Python — basic chat

pip install openai  # DeepSeek uses OpenAI SDK

from openai import OpenAI

client = OpenAI(
    api_key="your-deepseek-api-key",
    base_url="https://api.deepseek.com"
)

response = client.chat.completions.create(
    model="deepseek-chat",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Explain MoE architecture in 3 sentences."}
    ],
    temperature=0.7,
    max_tokens=512
)

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

Streaming response

stream = client.chat.completions.create(
    model="deepseek-chat",
    messages=[{"role": "user", "content": "Write a Python quicksort."}],
    stream=True
)

for chunk in stream:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)

Function calling (tool use)

tools = [{
    "type": "function",
    "function": {
        "name": "get_weather",
        "description": "Get current weather for a city",
        "parameters": {
            "type": "object",
            "properties": {
                "city": {"type": "string", "description": "City name"}
            },
            "required": ["city"]
        }
    }
}]

response = client.chat.completions.create(
    model="deepseek-chat",
    messages=[{"role": "user", "content": "What's the weather in Tokyo?"}],
    tools=tools,
    tool_choice="auto"
)

tool_call = response.choices[0].message.tool_calls[0]
print(tool_call.function.name)       # get_weather
print(tool_call.function.arguments)  # {"city": "Tokyo"}

5. Migration: Deprecated Endpoints (⚠️ July 24 Deadline)

Which endpoints are being deprecated?

Migration script (scan your codebase)

# Scan all .py files for deprecated model names
import os, re

DEPRECATED = [
    "deepseek-chat-v3",
    "deepseek-chat-v3-0324",
    "deepseek-reasoner-r1",
    "deepseek-reasoner-r1-0528",
]

def scan_dir(path):
    for root, _, files in os.walk(path):
        for f in files:
            if f.endswith(".py"):
                fpath = os.path.join(root, f)
                content = open(fpath).read()
                for dep in DEPRECATED:
                    if dep in content:
                        print(f"FOUND: {dep} in {fpath}")

scan_dir(".")  # run from your project root

Bulk replace with sed (Linux/macOS)

# Replace deprecated chat model
find . -name "*.py" -exec sed -i \
  's/deepseek-chat-v3[^"'"'"']*/deepseek-chat/g' {} \;

# Replace deprecated reasoner model
find . -name "*.py" -exec sed -i \
  's/deepseek-reasoner-r1[^"'"'"']*/deepseek-reasoner/g' {} \;

6. Run DeepSeek Locally with Ollama

For privacy-sensitive workloads, offline use, or zero API cost experimentation, you can run DeepSeek models locally. The distilled variants are small enough for consumer hardware.

Hardware requirements

Setup with Ollama

# Install Ollama (macOS/Linux)
curl -fsSL https://ollama.ai/install.sh | sh

# Pull and run DeepSeek R1 7B
ollama pull deepseek-r1:7b
ollama run deepseek-r1:7b

# Use via OpenAI-compatible API (Ollama exposes port 11434)
from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:11434/v1",
    api_key="ollama"  # any non-empty string
)

response = client.chat.completions.create(
    model="deepseek-r1:7b",
    messages=[{"role": "user", "content": "Hello!"}]
)
print(response.choices[0].message.content)

7. Building AI Agents with DeepSeek V4

DeepSeek V4 supports all the primitives needed for agentic applications: tool calling, structured output, long context, and streaming. Here's a minimal agent loop:

import json
from openai import OpenAI

client = OpenAI(
    api_key="your-deepseek-api-key",
    base_url="https://api.deepseek.com"
)

# Define tools
tools = [
    {
        "type": "function",
        "function": {
            "name": "web_search",
            "description": "Search the web for current information",
            "parameters": {
                "type": "object",
                "properties": {
                    "query": {"type": "string"}
                },
                "required": ["query"]
            }
        }
    }
]

def run_agent(user_input: str, max_turns: int = 5):
    messages = [{"role": "user", "content": user_input}]

    for turn in range(max_turns):
        response = client.chat.completions.create(
            model="deepseek-chat",
            messages=messages,
            tools=tools,
            tool_choice="auto"
        )
        
        msg = response.choices[0].message
        messages.append(msg)

        # No tool call = final answer
        if not msg.tool_calls:
            return msg.content

        # Execute each tool call
        for tc in msg.tool_calls:
            args = json.loads(tc.function.arguments)
            # Replace with your actual tool implementation
            result = f"[Search result for: {args['query']}]"
            messages.append({
                "role": "tool",
                "tool_call_id": tc.id,
                "content": result
            })

    return "Max turns reached"

answer = run_agent("What are the top AI agent frameworks in 2026?")
print(answer)

Using DeepSeek V4 with LangChain

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    model="deepseek-chat",
    api_key="your-deepseek-api-key",
    base_url="https://api.deepseek.com",
    temperature=0
)

# Works with all LangChain agents, chains, and tools
from langchain.agents import AgentExecutor, create_tool_calling_agent
from langchain_core.prompts import ChatPromptTemplate

prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a helpful assistant."),
    ("user", "{input}"),
    ("placeholder", "{agent_scratchpad}")
])

agent = create_tool_calling_agent(llm, tools=[], prompt=prompt)
executor = AgentExecutor(agent=agent, tools=[], verbose=True)
result = executor.invoke({"input": "Summarize DeepSeek V4 features"})

8. Limitations to Know Before You Commit

• No vision support in deepseek-chat. If you need image understanding, use GPT-4o or Gemini.
• No audio/video modalities — text only.
• No official SLA for uptime. For mission-critical apps, consider a fallback provider via LiteLLM or Portkey.
• Rate limits are lower than OpenAI's commercial tiers — check platform.deepseek.com for current limits.
• Latency can be higher than GPT-4o on complex requests due to MoE routing overhead.
• Data residency: requests are processed on DeepSeek's infrastructure in China. For GDPR-sensitive EU data, consider Mistral or on-premise deployment instead.

response = litellm.completion(
    model="deepseek/deepseek-chat",
    messages=messages,
    fallbacks=["gpt-4o-mini", "claude-haiku-3-5"]
)

9. Verdict: When to Use DeepSeek V4

Bottom line: DeepSeek V4 is the best-value LLM API available in 2026 for text-based tasks. If your workload doesn't need vision, a 1M+ context window, or EU data residency, there's almost no reason to pay 10x more for GPT-4o on equivalent quality.

Find DeepSeek, LiteLLM, Portkey, and 420+ other AI agent tools at AgDex.ai — the most comprehensive AI tools directory for developers.