What version of the Codex App are you using (From “About Codex” dialog)?
Version 26.721.41059
What subscription do you have?
plus
What platform is your computer?
windows 11
What issue are you seeing?
Subject: Codex custom agent role fails with unsupported gpt-5.6 model on ChatGPT account
I am using Codex with a ChatGPT account. In my local project, I have custom agent role profiles under .codex/agents/*.toml.
When a role profile contains:
model = "gpt-5.6"
spawning that agent fails before the agent starts.
Observed error:
400 invalid_request_error
"The 'gpt-5.6' model is not supported when using Codex with a ChatGPT account."
Example role file:
.codex/agents/gorev.toml
name = "gorev"
model = "gpt-5.6"
model_reasoning_effort = "medium"
Expected behavior:
Either Codex should accept the model if it is exposed in the UI/tooling, or the role configuration should clearly reject/validate it before spawn with a documented supported model list for ChatGPT accounts.
Actual behavior:
The agent spawn request reaches the API and fails with 400 invalid_request_error. The agent never starts.
Workaround:
Changing the same role profile to:
model = "gpt-5.5"
allows the agent to spawn successfully.
Impact:
Custom agent roles can silently break if a role profile uses gpt-5.6. The error is confusing because the failure happens at agent creation time, before the role prompt or task instructions are read.
Please clarify whether gpt-5.6 should be supported for Codex custom agent roles on ChatGPT accounts, or whether the supported model list for .codex/agents/*.toml should exclude it.
What steps can reproduce the bug?
Steps to reproduce:
-
Use Codex with a ChatGPT account.
-
In a local Codex project, create or edit a custom agent role file:
.codex/agents/gorev.toml
-
Set the role model to gpt-5.6:
name = "gorev"
description = "Test custom role"
model = "gpt-5.6"
model_reasoning_effort = "medium"
developer_instructions = '''
Test role. Do not modify files.
'''
-
Start/spawn this custom role from Codex.
Expected result:
The custom agent role should start, or Codex should reject the model earlier with a clear supported-model validation message.
Actual result:
The agent fails before starting with:
400 invalid_request_error
"The 'gpt-5.6' model is not supported when using Codex with a ChatGPT account."
Workaround:
Changing the same role file to model = "gpt-5.5" allows the custom role to start.
Session/context details:
Session ID: [add your Codex session/thread ID if available]
Token usage: not available to me
Context window usage: not available to me
Relevant project files tested:
.codex/agents/gorev.toml
.codex/agents
What is the expected behavior?
No response
Additional information
No response
What version of the Codex App are you using (From “About Codex” dialog)?
Version 26.721.41059
What subscription do you have?
plus
What platform is your computer?
windows 11
What issue are you seeing?
Subject: Codex custom agent role fails with unsupported gpt-5.6 model on ChatGPT account
I am using Codex with a ChatGPT account. In my local project, I have custom agent role profiles under
.codex/agents/*.toml.When a role profile contains:
model = "gpt-5.6"
spawning that agent fails before the agent starts.
Observed error:
400 invalid_request_error
"The 'gpt-5.6' model is not supported when using Codex with a ChatGPT account."
Example role file:
.codex/agents/gorev.toml
name = "gorev"
model = "gpt-5.6"
model_reasoning_effort = "medium"
Expected behavior:
Either Codex should accept the model if it is exposed in the UI/tooling, or the role configuration should clearly reject/validate it before spawn with a documented supported model list for ChatGPT accounts.
Actual behavior:
The agent spawn request reaches the API and fails with 400 invalid_request_error. The agent never starts.
Workaround:
Changing the same role profile to:
model = "gpt-5.5"
allows the agent to spawn successfully.
Impact:
Custom agent roles can silently break if a role profile uses
gpt-5.6. The error is confusing because the failure happens at agent creation time, before the role prompt or task instructions are read.Please clarify whether
gpt-5.6should be supported for Codex custom agent roles on ChatGPT accounts, or whether the supported model list for.codex/agents/*.tomlshould exclude it.What steps can reproduce the bug?
Steps to reproduce:
Use Codex with a ChatGPT account.
In a local Codex project, create or edit a custom agent role file:
.codex/agents/gorev.toml
Set the role model to gpt-5.6:
name = "gorev"
description = "Test custom role"
model = "gpt-5.6"
model_reasoning_effort = "medium"
developer_instructions = '''
Test role. Do not modify files.
'''
Start/spawn this custom role from Codex.
Expected result:
The custom agent role should start, or Codex should reject the model earlier with a clear supported-model validation message.
Actual result:
The agent fails before starting with:
400 invalid_request_error
"The 'gpt-5.6' model is not supported when using Codex with a ChatGPT account."
Workaround:
Changing the same role file to
model = "gpt-5.5"allows the custom role to start.Session/context details:
Session ID: [add your Codex session/thread ID if available]
Token usage: not available to me
Context window usage: not available to me
Relevant project files tested:
.codex/agents/gorev.toml
.codex/agents
What is the expected behavior?
No response
Additional information
No response