DeepSeek + Codex Complete Guide 2026: One-Click Setup to Use DeepSeek in Codex CLI, ChatGPT Desktop, and VS Code

「What if I could use DeepSeek with Codex (OpenAI's AI coding assistant)?」

In 2026, that dream became reality. The DeepSeek API natively supports Codex's Responses API, and with a one-click setup script, you can use DeepSeek models across Codex CLI, the ChatGPT desktop app, and the VS Code extension — all at once.

This article explains the DeepSeek official docs (api-docs.deepseek.com) in a beginner-friendly, visual way.

What you'll learn in this article

The basics of DeepSeek × Codex integration (what it can do)
Supported models (deepseek-v4-flash)
One-click setup (recommended) steps
Manual configuration (config.toml / models.json)
How to use it in Codex CLI, ChatGPT desktop, and VS Code
Session history caveats

Bottom line: what can you do?

You can swap the model in Codex (OpenAI's AI coding assistant) to DeepSeek.

Codex CLI, ChatGPT desktop app, and VS Code extension share the same config file
Configure once, and DeepSeek works in all three
Currently supported: deepseek-v4-flash (deepseek-v4-pro expected in early August 2026)
With an API key, no ChatGPT account is needed

Supported models

Model	Codex support	Notes
deepseek-v4-flash	✅ Supported	Fastest and lowest cost. Best for agentic coding
deepseek-v4-pro	🔜 Expected early Aug 2026	Highest-performance model

Two setup options

Method	Difficulty	Recommendation
Option 1: One-click setup script	★☆☆ (easy)	✅ Recommended
Option 2: Edit config files manually	★★☆ (intermediate)	For fine-grained control

Option 1: One-click setup (recommended)

Prerequisites

Install Codex CLI or the ChatGPT desktop app and launch it at least once (so the ~/.codex directory is created).

Steps

macOS / Linux (run in terminal):

bash <(curl -fsSL https://cdn.deepseek.com/api-docs/codex-deepseek-setup-en.sh)

Windows (run in PowerShell):

irm https://cdn.deepseek.com/api-docs/codex-deepseek-setup-en.ps1 | iex

After running

Select the model you want from the menu
On first run, enter your API key (starts with sk-)
- Get one at DeepSeek Platform
Done!

What the script does automatically

Step	Content
1. Backup	Saves existing ~/.codex/config.toml to ~/.codex/backup-deepseek/ (restorable anytime)
2. Model catalog	Writes DeepSeek model metadata to ~/.codex/models.json
3. config.toml edit	Rewrites only necessary fields + adds [model_providers.deepseek]. Existing MCP servers etc. are all preserved
4. Validate	Validates config.toml / models.json syntax. Aborts without changes on failure

Setup flow diagram

Option 2: Manual config file editing

1. Create the model catalog

Create ~/.codex/models.json (declares DeepSeek model metadata, including both deepseek-v4-flash and deepseek-v4-pro):

{
  "models": [
    {
      "slug": "deepseek-v4-flash",
      "context_window": 1048576,
      "max_context_window": 1048576,
      "effective_context_window_percent": 95,
      "display_name": "DeepSeek-V4-Flash",
      "default_reasoning_level": "high",
      "supported_reasoning_levels": [
        { "effort": "low", "description": "Fast responses with lighter reasoning" },
        { "effort": "high", "description": "Extra high reasoning depth for complex problems" },
        { "effort": "max", "description": "Maximum reasoning depth for the hardest problems" }
      ]
    }
  ]
}

2. Edit config.toml

Add the following to ~/.codex/config.toml:

model = "deepseek-v4-flash"
model_provider = "deepseek"
model_catalog_json = "~/.codex/models.json"

[model_providers.deepseek]
name = "deepseek"
base_url = "https://api.deepseek.com/"
wire_api = "responses"
experimental_bearer_token = "<your DeepSeek API Key>"

config.toml field reference

Field	Description
model	Default model to use
model_provider	Provider ID (must match [model_providers.<id>])
model_catalog_json	Path to the custom model catalog (models.json)
base_url	DeepSeek API endpoint
wire_api	Protocol; "responses" = Responses API
experimental_bearer_token	Your API key (stored directly in the config file)

Getting started

Once configured, DeepSeek works in all three clients.

Codex CLI

cd /path/to/my-project
codex

If the startup banner shows model: deepseek-v4-flash, the configuration is active!

ChatGPT desktop app

Mac: the model picker shows Custom (the DeepSeek model you selected is actually in use)
Windows: shows Custom or DeepSeek-V4-Flash

Codex IDE extension (VS Code)

Shares the same config as Codex CLI. Just install the extension and start using it.

Session history caveat

If previous sessions seem missing after switching providers, nothing is deleted.

Codex stores session history in separate groups by login method
Official ChatGPT subscription sessions and third-party API (DeepSeek) sessions are kept apart
Only the group matching the current config is shown
Restoring the previous config brings earlier sessions back
Restart the ChatGPT client after switching for the change to take effect

Summary

DeepSeek × Codex integration lets you do cutting-edge agentic coding at low cost without relying on expensive ChatGPT subscriptions.

One-click script sets everything up (with backup and validation)
Works in Codex CLI / ChatGPT desktop / VS Code extension
Supported model: deepseek-v4-flash (pro coming early August)
Session history is safely kept separate

Run the one-click script and try DeepSeek models in Codex today.

👉 Official docs: DeepSeek API Docs — Integrate with Codex 👉 Get an API key: DeepSeek Platform