Configure Claude Desktop to Use DeepSeek‑V4 Without Login or Subscription

Why This Combination Works

The Claude desktop client (now called Claude Cowork) provides two core agents: Cowork , which can read files, organize spreadsheets, and draft reports, and Code , a graphical interface that can run system commands and invoke skills. By running these agents locally, you get a smoother experience than the web version.

DeepSeek‑V4 is the newly released open‑source large model from DeepSeek. It comes in two variants:

V4‑Flash – fast and economical; input cost 0.14 CNY per million tokens, output 0.28 CNY.

V4‑Pro – high‑performance, comparable to top‑tier closed models; normally 3.48 CNY per million tokens but currently offered at a limited‑time 2.5× discount, i.e., 0.025 CNY per million tokens (cache‑hit price).

New DeepSeek accounts receive 5 million free tokens, which is more than enough for extended experimentation.

DeepSeek’s API is natively compatible with the Anthropic API protocol, allowing Claude Desktop to connect via its Gateway mode without any proxy or intermediate layer.

Configuration Steps

Download Claude Desktop from claude.com/download . It supports macOS 13+ and Windows 10/11. Install it and do not log in – keep the client in an unauthenticated state.

⚠️ If you have already logged in, log out before proceeding.

Enable Developer Mode (hidden deep in the menu): Help → Troubleshooting → Enable Developer Mode . After enabling, a new Developer menu appears.

Open the Developer menu and select Configure Third‑Party Inference… . In the dialog fill in the following fields:

Use this configuration : ✅ (must be on)

Gateway : select Anthropic‑compatible

Gateway base URL : https://api.deepseek.com/anthropic Gateway API key : your DeepSeek API key (see step 4)

Gateway auth scheme : bearer (required)

Obtain a DeepSeek API key: go to platform.deepseek.com , register a new account, navigate to API Keys , and create a new key. A small top‑up of 10–20 CNY is enough for long‑term use.

In the same configuration window, add a model to the Model list : <code>deepseek-v4-pro[1m]</code> The [1m] suffix unlocks the 1‑million‑token context window; be sure to enable the 1M‑context switch for the model.

💡 Use V4‑Pro instead of V4‑Flash for Cowork and Code because those agents trigger the model’s maximum reasoning intensity, effectively giving you the full‑power V4‑Pro Max.

Click Apply locally and then Relaunch Now . The client restarts, and you will see the Cowork and Code tabs with the default model set to deepseek-v4-pro[1m] . Send a test prompt; a response confirms successful configuration.

Common Issues

Q1: Cannot find the Developer menu

Make sure you have enabled it via Help → Troubleshooting → Enable Developer Mode.

Fully quit Claude Desktop and reopen it.

Ensure you are using the latest version of Claude Desktop.

Q2: Still redirected to the regular Claude login page

Confirm that Use this configuration is turned on.

After clicking Apply locally , completely exit and relaunch Claude Desktop.

If the problem persists, check the configuration report under Help → Troubleshooting.

Q3: Errors or cannot connect

Verify the API key is copied without extra spaces.

Ensure the base URL is exactly https://api.deepseek.com/anthropic (note the trailing /anthropic, not /v1).

Make sure the auth scheme is set to bearer .

Check that your DeepSeek account has sufficient balance.

Q4: Why does an OpenAI‑style endpoint fail?

Claude Desktop expects an Anthropic‑compatible endpoint. Use the /anthropic path; DeepSeek supports both formats, but the Anthropic one is required here.

Q5: How to switch back to the official Claude service?

Disable Use this configuration , fully quit Claude Desktop, and reopen it; the official login screen will appear.

Q6: Can free users access Cowork?

The third‑party mode does not depend on Claude Pro credits; the model calls go through your own DeepSeek API key, and the 5 million free tokens are more than enough.

Extending to Other Models

The Gateway accepts any Anthropic‑compatible model, so you can add, for example:

DeepSeek V4‑Pro (high performance)

DeepSeek V4‑Flash (fast, cheap)

Claude Sonnet 4.5 (if your API service supports it)

GPT‑5 (if you have access)

This flexibility lets a single desktop client switch among multiple models without buying a subscription plan.

Quick Recap

Download Claude Desktop and stay unlogged .

Enable Developer Mode .

Open Configure Third‑Party Inference .

Enter the DeepSeek gateway URL https://api.deepseek.com/anthropic, your API key, and select bearer authentication.

Add the model deepseek-v4-pro[1m] and turn on the 1M‑context switch.

Click Apply locally and restart the client.

After setup you can use Claude Cowork to read files, organize tables, and write reports, and Claude Code for programming tasks, all without spending any Claude subscription money.

New DeepSeek accounts receive 5 million free tokens. Even after they are exhausted, V4‑Flash remains extremely cheap at 0.14 CNY per million input tokens and 0.28 CNY per million output tokens.