Can't use other provider model

i have same problem, providers like openrouter or others doesnt workin

● OpenAI Codex (research preview) v0.1.2505161800            │
╰──────────────────────────────────────────────────────────────╯
╭──────────────────────────────────────────────────────────────╮
│ localhost session: b555630e7ba148338c14078db0b8bea1          │
│ ↳ workdir: ~/.../OneDrive/Aplikacje/mess/mess v2             │
│ ↳ model: grok-3-beta                                         │
│ ↳ provider: xai                                              │
│ ↳ approval: suggest  

I am also stuck with same problem, can't use gemini in openai.

I am also stuck with same problem, can't use gemini in openai.

Yup, although for me it says error 400, instead of 401.

same:

 codex --provider gemini -m gemini-2.0-flash
╭──────────────────────────────────────────────────────────────╮
│ ● OpenAI Codex (research preview) v0.1.2505171619            │
╰──────────────────────────────────────────────────────────────╯
╭──────────────────────────────────────────────────────────────╮
│ localhost session: d3b205271ad443ce8a7f37c20498895f          │
│ ↳ workdir: ~/workspace/test-folder                           │
│ ↳ model: gemini-2.0-flash                                    │
│ ↳ provider: gemini                                           │
│ ↳ approval: suggest                                          │
╰──────────────────────────────────────────────────────────────╯
user
das

    system
    ⚠️  OpenAI rejected the request. Error details: Status: 400, Code: unknown, Type: unknown, Message: 400 status code (no body). Please verify your settings and try again.
╭──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│                                                                                                                                                                                          │
╰──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
  ctrl+c to exit | "/" to see commands | enter to send — 100% context left```

Same issue.

!Image

Here's a solution that works for me.

Instead of setting your OPENROUTER_BASE_URL and OPENROUTER_API_KEY, set OPENAI_BASE_URL and OPENAI_API_KEY.

Example

export OPENAI_BASE_URL="https://openrouter.ai/api/v1"
export OPENAI_API_KEY="sk-***-21"

Note: In the above example, pay attention to the actual URL, I'm using "openrouter" endpoint and NOT "openai" endpoint.

Here's a solution that works for me. Instead of setting your OPENROUTER_BASE_URL and OPENROUTER_API_KEY, set OPENAI_BASE_URL and OPENAI_API_KEY. Example export OPENAI_BASE_URL="https://openrouter.ai/api/v1" export OPENAI_API_KEY="sk-***-21" Note: In the above example, pay attention to the actual URL, I'm using "openrouter" endpoint and NOT "openai" endpoint.

this work for me:
Currently, if you want to use DeepSeek, you can configure it as follows:

1. set env:

export OPENAI_BASE_URL=https://api.deepseek.com/
export OPENAI_API_KEY=sk-****15
export DEEPSEEK_API_KEY=sk-****15

1. change model and provider:

~/.codex/config.json:

``` json

{
"model": "deepseek-chat",
"provider": "deepseek",
...
}

3. run： `codex --approval-mode full-auto "create the fanciest todo-list app"`

!Image
I have the same problem, it seems like I have no support for function calls

@pallavbakshi Thanks for the trick, can you check if you can use this for gemini. I tried to use the url mention in repo as openai which end up as 404, I tried some other way, still not working

Same here. Gemini provider not working. A regular Curl call shows there is no problem with Google's OpenAI API:

curl "https://generativelanguage.googleapis.com/v1beta/openai/chat/completions" -H "Content-Type: application/json" -H "Authorization: Bearer $GEMINI_API_KEY" -d '{
    "model": "gemini-2.5-pro-preview-03-25",
    "messages": [
        {"role": "user", "content": "Explain to me how AI works"}
    ],
    "stream": true
  }'
...
data: {"choices":[{"delta":{"content":"Okay, let's break down how Artificial Intelligence (AI) works, starting simple and then adding a","role":"assistant"},"index":0}],"created":1747711087,"model":"gemini-2.5-pro-preview-03-25","object":"chat.completion.chunk"}

data: {"choices":[{"delta":{"content":" bit more detail.\n\nImagine you're trying to teach a child to recognize a cat.\n\n1.  **","role":"assistant"},"index":0}],"created":1747711087,"model":"gemini-2.5-pro-preview-03-25","object":"chat.completion.chunk"}
...

Codex just fails with ambiguous 400 error:

user
provide an overview of the current project
╭────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│(   ●  ) 1s  Thinking..                                                                                                                                        press Esc twice to interrupt │
╰────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
  ctrl+c to exit | "/" to see commands | enter to send — 100% context left

OpenAI:DEBUG:response (error; (error; no more retries left)) 400 https://generativelanguage.googleapis.com/v1beta/openai/chat/completions {
  'alt-svc': 'h3=":443"; ma=2592000,h3-29=":443"; ma=2592000',
  'content-encoding': 'gzip',
  'content-type': 'application/json; charset=UTF-8',
  date: 'Tue, 20 May 2025 03:16:35 GMT',
  server: 'scaffolding on HTTPServer2',
  'server-timing': 'gfet4t7; dur=876',
  'transfer-encoding': 'chunked',

    system
    ⚠️  OpenAI rejected the request. Error details: Status: 400, Code: unknown, Type: unknown, Message: 400 status code (no body). Please verify your settings and try again.

just rm ~/.codex/auth.json
It's works for me.

There's no auth.json here

it worked for me like this:

export GEMINI_API_KEY="your gemini key here"
export OPENAI_API_KEY="$GEMINI_API_KEY"
export GEMINI_BASE_URL="https://generativelanguage.googleapis.com/v1beta/openai/"
codex -p gemini -m gemini-2.5-flash-preview-05-20

just rm ~/.codex/auth.json It's works for me.

thanks!!!!

just rm ~/.codex/auth.json It's works for me.

when I removed the json, I use following command.

export DEEPSEEK_API_KEY="sk-XXXXXX"

export OPENAI_API_KEY="$DEEPSEEK_API_KEY"

codex -p deepseek -m deepseek-chat

I removed the auth.json and tried to write "hello world" it's end up as this error, any idea?

!Image

cc @bedorlan

I removed the auth.json and tried to write "hello world" it's end up as this error, any idea?

yes, im getting that one too when using the gemini-2.5-flash model.
it has been reported here: https://github.com/openai/codex/issues/1077

im just using the gemini-2.5-pro for now.
(the flash model is not good atm anyway 🙈, at least for what i need)

@bedorlan Just to confirm, Gemini 2.5 Pro is not available on the free tier. Google discontinued the pro-exp-03-xx model a few weeks ago. I’ll try using Azure OpenAI instead — I noticed a positive case mentioned in this comment.

@anirugu i had success using gpt-4.1-mini, that one is what works best for me cost-effective.

@bedorlan I can confirm that Azure OpenAI is working well for me. It took significantly less effort to get it running compared to Google Gemini. Thanks!

yes, im getting that one too when using the gemini-2.5-flash model. it has been reported here: #1077

I have a fix for it here: https://github.com/openai/codex/pull/1079

But I'm actually encountering the error with azure o3. Are you guys not seeing the error with azure gpt-4.1-mini?

<img width="682" alt="Image" src="https://github.com/user-attachments/assets/93e5def0-8aba-4b1d-adeb-b51e5ec6b73a" />

Facing the same issue. Kinda pathetic tbh

You might find this branch helpful: https://github.com/msmhrt/codex/tree/workaround/gemini-bugfix

just rm ~/.codex/auth.json It's works for me.

Do it inside directory that you are using codex.
Thank you,

It's been a problem for a long time. It's never been fixed.

@FocusProgram I wonder if we can just fork it and ask Copilot to fix the issue in the background 😂

just rm ~/.codex/auth.json It's works for me.

awesome~ u save my day, my job and my life <3

My solution for this problem:

export GEMINI_API_KEY="qwertyuiop12345"
export OPENAI_API_KEY="$GEMINI_API_KEY"

codex --provider gemini

As i understand, if you don't set the OPENAI_API_KEY variable, codex simply won't start

And yes, OPENAI_API_KEY should also be the Gemini API key.

Here is what I did to use deepseek with codex:

I made a /.codex/config.toml in the root of my project that looked like this

model = "deepseek-chat"
model_provider = "deepseek"

[model_providers.deepseek]
name     = "DeepSeek"
base_url = "https://api.deepseek.com"
env_key  = "DEEPSEEK_API_KEY"
stream = true

Then I added export CODEX_HOME="$(pwd)/.codex" to my ~/.bashrc file so that codex will know exactly where to find the config file. I also exported my deepseek key from the .bashrc file, but I think that is optional if you prefer to export it each time you want to run codex, and that was it. All I have to do in the future is just add the config.toml file and codex should run smoothly.

For additional context, brew install codex will install rust-based codex that only works with .toml files and ignores .json files.

Edit: I use Ububtu 22.04.5 in wsl2 on windows so I only just learned that the ~/.codex had been in the user root (not sure if that's how it's called) in wsl all along. I just didn't see it because I did not ls with the -a flag, which I also just learned about, so I ended up thinking I had to create it manually in the project's root. I'm still learning the ropes so my apologies to all the more experienced people that may have gotten confused reading my comment.

Thank you for the feedback! Please try again with the latest version of Codex CLI. If the issue persists, we would appreciate a report on the codex-rs/ implementation. You can find more details about the project update here: https://github.com/openai/codex/discussions/1174.

unexpected status 404 Not Found: Unknown error, url: https://api.deepseek.com/responses
codex deepseek

For people still landing here from DeepSeek searches: DeepSeek V4 still exposes the OpenAI-compatible Chat Completions surface, not a native /responses endpoint, so direct https://api.deepseek.com/responses / /v1/responses fails.

We published the local shim we are using to run DeepSeek V4 from current Codex:

https://github.com/holo-q/deepseek-responses-proxy

It gives Codex a local Responses endpoint and forwards to DeepSeek /chat/completions. Tested with Codex 0.128.0 using both deepseek-v4-pro and deepseek-v4-flash.

Minimal current config shape:

[model_providers.deepseek]
name = "DeepSeek"
base_url = "http://127.0.0.1:8787/v1"
experimental_bearer_token = "codex-deepseek-local"
wire_api = "responses"

[profiles.deepseek-v4-pro]
model_provider = "deepseek"
model = "deepseek-v4-pro"
model_context_window = 1000000
approval_policy = "untrusted"
sandbox_mode = "workspace-write"

Start Codex with the profile, not just the model flag:

codex -p deepseek-v4-pro

Using codex -m deepseek-v4-pro by itself leaves the provider on the default OpenAI/ChatGPT account path, which causes the misleading “model is not supported when using Codex with a ChatGPT account” error.