Official preview release · Web, app, and API

DeepSeek V4: Future-Ready Reasoning for Teams

DeepSeek-V4 is now available through chat.deepseek.com, the DeepSeek app, and the DeepSeek API. The current API lineup includes DeepSeek-V4-Pro and DeepSeek-V4-Flash with 1M context support and OpenAI/Anthropic-compatible endpoints.

Community Demo

Try a live DeepSeek V4 community demo here, and use chat.deepseek.com, platform.deepseek.com, or the DeepSeek API docs when you need the official product surfaces.

What is DeepSeek V4?

DeepSeek-V4 is an officially released preview family announced by DeepSeek in April 2026. The release includes two API models, DeepSeek-V4-Pro and DeepSeek-V4-Flash.

Official API Model Names

DeepSeek documents two current API model names: deepseek-v4-pro and deepseek-v4-flash. The older aliases deepseek-chat and deepseek-reasoner still work for compatibility today, but the public docs mark them for deprecation on July 24, 2026.

1M Context and Thinking Modes

Both V4-Pro and V4-Flash are listed with 1M context length, a maximum output of 384K, and support for both thinking and non-thinking modes in the official pricing table.

OpenAI and Anthropic Compatible

DeepSeek keeps public base URLs for both OpenAI-format and Anthropic-format access, which means many existing SDK flows can be adapted without a custom client.

Tooling and Structured Output

The public model table lists JSON output, tool calls, chat prefix completion, and beta FIM completion support. FIM is documented for non-thinking mode only.

Why Choose DeepSeek V4?

The official release focuses on long context, agentic coding, and lower-cost API access rather than vague roadmap promises.

Two-Tier Model Lineup

Use V4-Flash for lower-cost, higher-throughput calls, or V4-Pro when you want the stronger premium model in the same family.

Fact-Based Integration

You can use a standard OpenAI SDK, point it at api.deepseek.com, and change the model name without inventing a custom client layer.

Public Pricing and Limits

DeepSeek publishes per-1M-token pricing, cache-hit discounts, context length, max output, and concurrency limits in the public docs.

How to Integrate DeepSeek V4

The official quick start is simple: create an API key, keep the DeepSeek base URL, then call deepseek-v4-pro or deepseek-v4-flash from an OpenAI-compatible client.

  • Create an API key

    Generate a key in platform.deepseek.com and use https://api.deepseek.com as the OpenAI-format base_url.

  • Choose the model

    Use deepseek-v4-pro for the premium tier or deepseek-v4-flash for the lower-cost tier. The docs also expose an Anthropic-format base URL at /anthropic.

  • Enable reasoning when needed

    The official examples show thinking enabled and reasoning_effort set explicitly, so you can raise reasoning depth only for tasks that need it.

from openai import OpenAI

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

response = client.chat.completions.create(
  model="deepseek-v4-pro",
  messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)

FAQ About DeepSeek V4

Use the Official DeepSeek Sources

For real access and current pricing, go to chat.deepseek.com, platform.deepseek.com, and api-docs.deepseek.com. This site summarizes those public facts and links back to the primary pages.