Deprecating chat/completions support in Codex

[etraut-openai]

In early 2025, OpenAI introduced the responses API — an interface optimized for reasoning models, multi-turn conversations, and tool-rich workflows. This API replaces the older chat/completions API, which originated in the GPT-3.5 era and was not designed for today’s agentic coding and reasoning use cases.

While Codex has historically supported both APIs, the chat/completions API has increasingly hampered our ability to improve Codex and deliver new features. Maintaining compatibility with this legacy protocol has added complexity, introduced regressions, and increased support overhead.

As a result, the Codex team is deprecating support for the chat/completions API. Full removal is slated for early February 2026.

What this means

• Codex feature requests related to the chat/completions API will be closed.
• During the deprecation period, bug reports involving the chat/completions API will be triaged based on user impact.
• The Codex CLI will be updated to emit a deprecation warning if configured to use the chat/completions API.
• In February 2026, this will transition to a hard error as support is fully removed.

Migrating from chat/completions to responses API

The vast majority of Codex users are unaffected by this change. If you are using OpenAI-hosted models with a default Codex configuration, you are already using the responses API, and no action is required.

If you have configured a custom model provider with wire_api = "chat" or have not specified a wire_api, you will need to update your model provider definition.

[model_providers.my_custom_provider]
wire_api = "responses"

Local models / OSS usage

For local models — including OpenAI’s gpt_oss, you can use LM Studio, which supports the responses API.

To configure Codex to use LM Studio:

oss_provider = "lmstudio"

Maintainers of Ollama are planning to add support for the responses API in the near future.

Enterprise Environments

If your organization uses an LLM proxy or gateway, please coordinate with your IT or platform team to ensure your proxy supports the responses API before the end of the deprecation period.

Why we’re making this change

This change allows Codex to:

• Deliver more reliable behavior and a better developer experience
• Accelerate feature development
• Reduce regressions and compatibility issues
• Align with current and future OpenAI platform standards

We recognize that migrations can be inconvenient and appreciate your understanding as we continue to evolve Codex to improve your development experience.

[gaoconggit]

Has support for the chat API been removed in version 0.84.0? It seems that wire_api = "chat" is no longer supported.

[etraut-openai]

Support for "chat" won't be removed until Feb 1. There is a deprecation notice in place.

It's possible that there was a regression introduced that breaks some implementations of "chat/completions". We've struggled to keep it working as we introduce new features. If you suspect a regression, please file a bug report and describe the use case. We'll prioritize fixes based on user impact. If it's a corner case (affects very few users), we'll likely leave it unfixed since the code is going away shortly.

[gaoconggit]

I'm using z.ai, version 0.80.0 works, but later versions don't. 😂

[gaoconggit]

[jemshit]

Just to repeat, you need to add wire_api under model_providers:

[model_providers.custom_provider]
wire_api = "responses"

Not profiles:

[profiles.model_name]
wire_api = "responses"

[gaoconggit]

I'm using z.ai, version 0.80.0 works, but later versions don't. 😂

[gaoconggit]

[gaoconggit]

It stopped working after the upgrade.

[gaoconggit]

[gaoconggit]

I think I've found the reason; it's because glm-4.7 is not supported.

{
"error": {
"code": "1214",
"message": "Incorrect role information"
}
}

[FVilli]

LM Studio 0.3.39 (latest) + codex-cli 0.92.0 does not work with wire_api = "responses"

I’m experiencing an incompatibility between LM Studio 0.3.39 and codex-cli 0.92.0 when using the new Responses API.

With the following config.toml, Codex fails with stream disconnection errors:

profile = "local-code"
oss_provider = "lmstudio"

[model_providers.lms]
name = "LM Studio"
type = "openai"
base_url = "http://monster.netbird.selfhosted:12345/v1"
api_key = "dummy"
wire_api = "responses"

[profiles.local-code]
model_provider = "lms"
model = "gpt-oss:120b"

Codex don't call:

POST /v1/responses

If I remove:

wire_api = "responses"

Codex falls back to the legacy Chat Completions API.
I get a deprecation warning, but everything works correctly and reliably.

Summary

• ✅ /v1/responses works via curl (non-streaming)
• ❌ Codex streaming via Responses API fails
• ✅ Legacy Chat API works fine

Is this a streaming incompatibility between LM Studio and Codex when using the Responses API ?

[etraut-openai]

I tested this about three weeks ago, and it worked then. That probably means there was either a regression in LMStudio or Codex since then or Codex introduced a change in the way that it uses the "responses" API that LMStudio doesn't properly handle.

If you're so inclined, you could try downgrading Codex and/or LMStudio to validate whether it's a regression.

[FedorenkoCodes]

Have the same issue with codex "0.92.0" with 2 versions of litellm - 1.75.8 and 1.80.15

With the following config

[model_providers.lite_llm]
name = "LiteLLM"
base_url = "https://gateway.azure.my_company.com"
env_key = "OPENAI_API_KEY"
wire_api = "chat"
env_http_headers = { "api-key" = "OPENAI_API_KEY" }

[profiles.lite_llm]
model_provider = "lite_llm"
model = "gpt-5.1-codex"

it chats, but never uses any tools.

But when I change wire_api = "responses" it plain doesn't connect

Both LiteLLM instances run "gpt-5.1-codex" with "mode": "responses"