DeepSeek Anthropic API: How to Use DeepSeek with the Anthropic SDK and Claude Code

Last updated: 4 May 2026

The DeepSeek Anthropic API lets developers call DeepSeek models through an Anthropic-compatible API format. In practice, this means you can point Anthropic-style tooling, including the Anthropic SDK and Claude Code, to DeepSeek’s Anthropic-compatible base URL: https://api.deepseek.com/anthropic. It does not mean DeepSeek is Anthropic, and it does not mean Anthropic hosts DeepSeek models. It means DeepSeek provides a compatibility layer for teams already using Anthropic-style workflows.

This guide explains the DeepSeek Anthropic API endpoint, SDK setup, Claude Code configuration, supported fields, unsupported fields, model choices, pricing notes, and troubleshooting steps.

Quick answer

Item	Answer
Endpoint	https://api.deepseek.com/anthropic
SDK	Anthropic Python SDK or TypeScript/JavaScript SDK
API key	Use your DeepSeek API key, not an Anthropic key
Recommended models	deepseek-v4-flash, deepseek-v4-pro
Best use case	Testing DeepSeek in Anthropic-style apps, Claude Code, agents, and migration workflows
Key limitation	Compatibility is not identical to Anthropic Claude API; images, documents, and several Claude-native content blocks are not supported in DeepSeek’s compatibility endpoint

What Is the DeepSeek Anthropic API?

The DeepSeek Anthropic API is DeepSeek’s Anthropic-compatible API format. Instead of rewriting an application that already uses client.messages.create() from the Anthropic SDK, you can change the base URL, use a DeepSeek API key, and call DeepSeek models with an Anthropic-style request shape.

DeepSeek’s official docs state that its API supports OpenAI/Anthropic-compatible formats and lists the Anthropic base URL as https://api.deepseek.com/anthropic. The same docs list the OpenAI-compatible base URL as https://api.deepseek.com.

Compatibility is useful when you want to:

• use DeepSeek with Anthropic SDK code;
• test DeepSeek as a backend for Claude Code;
• reduce migration work for existing Anthropic-style apps;
• compare DeepSeek models against Claude models in controlled workflows;
• keep a similar request structure while switching providers.

However, compatibility does not mean every Claude-native feature works. DeepSeek documents specific supported, ignored, and unsupported fields. For example, text content and tool use are supported, while image, document, search_result, server tool, web search tool, MCP tool, and container upload content blocks are not supported in the Anthropic-compatible endpoint.

DeepSeek Anthropic API Endpoint and Basic Setup

Field	Value	Notes
Anthropic-compatible base URL	https://api.deepseek.com/anthropic	Use this with the Anthropic SDK or Anthropic-style tools
OpenAI-compatible base URL	https://api.deepseek.com	Use this with OpenAI-compatible Chat Completions clients
API key	DeepSeek API key	Create it from the DeepSeek Platform
Model examples	deepseek-v4-flash, deepseek-v4-pro	Current model names listed in DeepSeek docs
SDK	anthropic Python SDK or @anthropic-ai/sdk	Anthropic SDKs provide message, streaming, and tool interfaces
Streaming	Supported	DeepSeek marks stream as fully supported
Tool calls	Supported	Tool name, input schema, and description are fully supported
JSON output	Supported on current DeepSeek V4 models	Verify current docs before production use

DeepSeek’s current model table lists deepseek-v4-flash and deepseek-v4-pro. It also says deepseek-chat and deepseek-reasoner are legacy names scheduled for deprecation on July 24, 2026, and currently map to non-thinking and thinking modes of deepseek-v4-flash.

How to Call DeepSeek with the Anthropic Python SDK

Install the Anthropic Python SDK:

python -m pip install anthropic

Set environment variables:

export ANTHROPIC_BASE_URL="https://api.deepseek.com/anthropic"
export ANTHROPIC_API_KEY="your_deepseek_api_key"

Use this Python example:

import os
from anthropic import Anthropic

DEEPSEEK_ANTHROPIC_BASE_URL = os.environ.get(
    "ANTHROPIC_BASE_URL",
    "https://api.deepseek.com/anthropic",
)

api_key = os.environ.get("ANTHROPIC_API_KEY")

if not api_key:
    raise RuntimeError(
        "Missing ANTHROPIC_API_KEY. Set it to your DeepSeek API key before running."
    )

client = Anthropic(
    api_key=api_key,
    base_url=DEEPSEEK_ANTHROPIC_BASE_URL,
)

try:
    message = client.messages.create(
        model="deepseek-v4-pro",
        max_tokens=1000,
        system="You are a helpful technical assistant.",
        messages=[
            {
                "role": "user",
                "content": "Explain the DeepSeek Anthropic API in one paragraph.",
            }
        ],
    )

    for block in message.content:
        if getattr(block, "type", None) == "text":
            print(block.text)

except Exception as exc:
    print(f"DeepSeek Anthropic-compatible request failed: {exc}")

DeepSeek’s own Anthropic API guide shows the same core pattern: install anthropic, set ANTHROPIC_BASE_URL, set an API key, create an anthropic.Anthropic() client, and call client.messages.create() with a DeepSeek model such as deepseek-v4-pro.

Anthropic’s Python SDK documentation confirms the package name is anthropic, supports synchronous and asynchronous usage, and uses the Messages API pattern with client.messages.create().

Node.js Example for the DeepSeek Anthropic-Compatible API

Install the Anthropic JavaScript SDK:

npm install @anthropic-ai/sdk

Set environment variables:

export ANTHROPIC_BASE_URL="https://api.deepseek.com/anthropic"
export ANTHROPIC_API_KEY="your_deepseek_api_key"

Create deepseek-anthropic-example.mjs:

import Anthropic from "@anthropic-ai/sdk";

const apiKey = process.env.ANTHROPIC_API_KEY;
const baseURL =
  process.env.ANTHROPIC_BASE_URL || "https://api.deepseek.com/anthropic";

if (!apiKey) {
  throw new Error("Missing ANTHROPIC_API_KEY. Set it to your DeepSeek API key.");
}

const client = new Anthropic({
  apiKey,
  baseURL,
});

try {
  const message = await client.messages.create({
    model: "deepseek-v4-pro",
    max_tokens: 1000,
    system: "You are a practical software engineering assistant.",
    messages: [
      {
        role: "user",
        content: "Give me a short checklist for using DeepSeek with the Anthropic SDK.",
      },
    ],
  });

  for (const block of message.content) {
    if (block.type === "text") {
      console.log(block.text);
    }
  }
} catch (error) {
  console.error("DeepSeek Anthropic-compatible request failed:", error);
}

Run it:

node deepseek-anthropic-example.mjs

Anthropic’s TypeScript SDK docs list npm install @anthropic-ai/sdk, server-side JavaScript usage, streaming support, and client.messages.create() as the standard request method.

How to Use DeepSeek in Claude Code

Claude Code is an AI coding assistant that runs in the terminal. DeepSeek documents an official integration path that points Claude Code to the DeepSeek Anthropic API endpoint.

For macOS or Linux:

export ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic
export ANTHROPIC_AUTH_TOKEN=<your DeepSeek API Key>
export ANTHROPIC_MODEL=deepseek-v4-pro[1m]
export ANTHROPIC_DEFAULT_OPUS_MODEL=deepseek-v4-pro[1m]
export ANTHROPIC_DEFAULT_SONNET_MODEL=deepseek-v4-pro[1m]
export ANTHROPIC_DEFAULT_HAIKU_MODEL=deepseek-v4-flash
export CLAUDE_CODE_SUBAGENT_MODEL=deepseek-v4-flash
export CLAUDE_CODE_EFFORT_LEVEL=max

For Windows PowerShell:

$env:ANTHROPIC_BASE_URL="https://api.deepseek.com/anthropic"
$env:ANTHROPIC_AUTH_TOKEN="<your DeepSeek API Key>"
$env:ANTHROPIC_MODEL="deepseek-v4-pro[1m]"
$env:ANTHROPIC_DEFAULT_OPUS_MODEL="deepseek-v4-pro[1m]"
$env:ANTHROPIC_DEFAULT_SONNET_MODEL="deepseek-v4-pro[1m]"
$env:ANTHROPIC_DEFAULT_HAIKU_MODEL="deepseek-v4-flash"
$env:CLAUDE_CODE_SUBAGENT_MODEL="deepseek-v4-flash"
$env:CLAUDE_CODE_EFFORT_LEVEL="max"

Then run:

cd /path/to/my-project
claude

Do not expose your DeepSeek API key in client-side code, screenshots, Git repositories, CI logs, or shared shell history.

DeepSeek Anthropic API Compatibility: Supported and Unsupported Fields

Area	Field or content type	Status
HTTP header	x-api-key	Fully supported
HTTP header	anthropic-beta, anthropic-version	Ignored
Simple field	model	Use DeepSeek model names
Simple field	max_tokens	Fully supported
Simple field	stop_sequences	Fully supported
Simple field	stream	Fully supported
Simple field	system	Fully supported
Simple field	temperature	Fully supported, range 0.0 to 2.0
Simple field	top_p	Fully supported
Simple field	thinking	Supported, but budget_tokens is ignored
Simple field	output_config	Only effort is supported
Simple field	container, mcp_servers, metadata, service_tier, top_k	Ignored
Tool field	tools.name	Fully supported
Tool field	tools.input_schema	Fully supported
Tool field	tools.description	Fully supported
Tool field	tools.cache_control	Ignored
Tool choice	none, auto, any, tool	Supported; disable_parallel_tool_use is ignored
Message content	string content	Fully supported
Message content	text block	Fully supported
Message content	image block	Not supported
Message content	document block	Not supported
Message content	search result block	Not supported
Message content	thinking block	Supported
Message content	redacted thinking block	Not supported
Message content	tool use / tool result	Supported
Message content	server tool, web search tool, code execution tool, MCP tool, container upload	Not supported

This matters when migrating Claude-native applications. A text-only chatbot may need very few changes. A Claude-native app that depends on image inputs, document blocks, web search tool results, server tools, or MCP content blocks will need compatibility testing and possibly a different integration path.

DeepSeek Models for Anthropic-Compatible Requests

DeepSeek currently lists two primary V4 model names for API use:

Model	Best for	Notes
deepseek-v4-flash	Lower-cost, high-volume, general-purpose usage	Also used for automatic mapping of unsupported model names in the Anthropic-compatible endpoint
deepseek-v4-pro	More complex coding, reasoning, and agentic workflows	Used in DeepSeek’s Anthropic SDK example and Claude Code configuration

DeepSeek’s pricing/model page lists both V4 models with support for thinking and non-thinking modes, 1M context length, maximum output up to 384K, JSON output, tool calls, and Chat Prefix Completion. FIM Completion is listed as non-thinking mode only. Check the official pricing/docs before production use because model features and prices may change.

DeepSeek’s Anthropic API guide also notes that unsupported model names passed to the DeepSeek Anthropic API are automatically mapped to deepseek-v4-flash. For production systems, treat this as a safety fallback, not as a substitute for explicit model selection and logging.

DeepSeek API Pricing Notes

DeepSeek lists pricing in units of 1M tokens and says billing is based on total input and output tokens. As checked in May 2026, the pricing page lists:

Model	1M input tokens, cache hit	1M input tokens, cache miss	1M output tokens
deepseek-v4-flash	$0.0028	$0.14	$0.28
deepseek-v4-pro	$0.003625 with listed discount	$0.435 with listed discount	$0.87 with listed discount

DeepSeek’s page says the deepseek-v4-pro discount is extended until May 31, 2026, 15:59 UTC, and also says product prices may vary, so you should regularly check the official pricing page before production use.

A simple cost formula is:

cost =
  (cache_hit_input_tokens / 1,000,000 × cache_hit_input_price)
+ (cache_miss_input_tokens / 1,000,000 × cache_miss_input_price)
+ (output_tokens / 1,000,000 × output_price)

For production, log token usage, cache-hit/cache-miss behavior, selected model name, and request type.

DeepSeek Anthropic API vs Anthropic Claude API

Category	DeepSeek Anthropic-compatible API	Anthropic Claude API
Provider	DeepSeek	Anthropic
Models	DeepSeek models such as deepseek-v4-flash and deepseek-v4-pro	Claude models
Endpoint	https://api.deepseek.com/anthropic	Anthropic’s native Claude API endpoints
SDK compatibility	Works with Anthropic-style SDK calls by changing configuration	Native SDK support
Native feature support	Compatibility layer with documented limitations	Full Claude-native feature access
Vision/documents	Not supported in DeepSeek’s Anthropic-compatible message content blocks	Supported in Claude-native workflows where available
Tool use	Supported for documented tool fields	Native Claude tool-use support
Pricing model	DeepSeek pricing	Anthropic pricing
Best for	Testing DeepSeek in Anthropic-style workflows, Claude Code, migration experiments	Claude-native applications and full Anthropic feature access
Migration risk	Medium if your app uses advanced Claude-native blocks	Lower for Claude-native apps

Anthropic’s own API docs describe official SDKs that handle authentication, request formatting, retries, streaming, timeouts, and error handling. They also emphasize that the direct Claude API is best for full Claude feature access.

Common Errors and Troubleshooting

Wrong base URL

Use:

https://api.deepseek.com/anthropic

Do not use the OpenAI-compatible base URL when calling through the Anthropic SDK.

Using Anthropic model names

Use DeepSeek model names such as:

deepseek-v4-flash
deepseek-v4-pro

Do not use Claude model names unless you are calling Anthropic directly.

Missing API key

For SDK examples, set:

export ANTHROPIC_API_KEY="your_deepseek_api_key"

For Claude Code, DeepSeek’s docs use:

export ANTHROPIC_AUTH_TOKEN="<your DeepSeek API Key>"

Unsupported image or document content

Remove image and document blocks before sending the request through DeepSeek’s Anthropic-compatible endpoint. DeepSeek documents image and document message content blocks as unsupported.

max_tokens errors

max_tokens is fully supported, but the generated output plus input must fit within the model’s limits. Cap max_tokens to a practical value for your application.

429 or rate limit errors

Add retry logic with exponential backoff, queue high-volume jobs, and monitor usage. Do not blindly retry all requests forever.

Streaming issues

Confirm stream is enabled only when your client code can consume server-sent events. Anthropic’s SDK docs show streaming support, and DeepSeek marks stream as fully supported in its compatibility table.

Unexpected automatic model mapping

If you pass an unsupported model name, DeepSeek says the Anthropic API backend automatically maps it to deepseek-v4-flash. Log the model you requested and the model behavior you observe, especially in production.

Production Best Practices

Keep API keys server-side. Never expose DeepSeek keys in browser bundles, public repositories, or mobile app binaries.

Use environment variables for all secrets and base URLs. This makes it easy to switch between DeepSeek, Anthropic, staging, and production environments.

Validate model outputs before using them in user-visible or business-critical workflows.

Validate tool-call arguments against your own schema before executing tools.

Add retry logic with exponential backoff for transient failures.

Log model name, latency, token usage, cache-hit/cache-miss data, and request outcome.

Cap max_tokens to control cost and reduce runaway outputs.

Use streaming for chat UX when appropriate.

Monitor official model and pricing changes.

Add fallback behavior for critical systems.

Test compatibility before migration, especially if your existing Claude workflow uses images, documents, citations, web search, server tools, code execution tool results, MCP tool calls, or advanced content blocks.

FAQ

Conclusion

The practical takeaway is simple: use the DeepSeek Anthropic API when you want DeepSeek models inside Anthropic-style workflows without rewriting your entire application. Set the base URL to https://api.deepseek.com/anthropic, use a DeepSeek API key, select deepseek-v4-flash or deepseek-v4-pro, and test every feature your app depends on.