Fall back to read 'reasoning_content' when 'reasoning' to support VLLM-served gpt-oss

[borishim]

What version of Codex is running?

codex-cli 0.34.0

Which model were you using?

gpt-oss-120b

What platform is your computer?

Darwin Sangwoos-MacBook-Pro-10.local 24.6.0 Darwin Kernel Version 24.6.0: Mon Jul 14 11:30:29 PDT 2025; root:xnu-11417.140.69~1/RELEASE_ARM64_T6000 arm64

What steps can reproduce the bug?

Serve gpt-oss-120b on vllm with the following options:

vllm serve openai/gpt-oss-120b \
--tensor-parallel-size 1 \
--tool-call-parser openai --reasoning-parser openai_gptoss --enable-auto-tool-choice \
--gpu-memory-utilization 0.95 \
--max-num-batched-tokens 1024

And put the following entries to ~/.codex/config.toml

[model_providers.vllm]
name = "OpenAI using Chat Completions"
base_url = "OUR_INTERNAL_URL"
env_key = "OPENAI_API_KEY"
wire_api = "chat"

[profiles.gpt-oss-120b-vllm]
model_provider = "vllm"
model = "openai/gpt-oss-120b"                                  

Then run codex with --profile=gpt-oss-120b-vllm.

codex answers simple questions, but does not respond to reasoning response, because with VLLM chat completion API, reasoning content is delivered with 'reasoning_content' key, but codex expect 'reasoning' key.

What is the expected behavior?

codex should work with VLLM-served gpt-oss-120b model, especially for reasoning and tool_call.

What do you see instead?

codex cannot support reasoning responses with chat completion.

Additional information

I will submit a PR to fix the behavior

[bartlettroscoe]

Given that #3722 is closed, should this issue not also be closed? See:

• Fall back to 'reasoning_content' to support vllm-served gpt-oss #3722 (comment)

?

[borishim]

That's right. I will close the isssue

[tylike]

ollama/llama.cpp/lmstudio all use reasoning_content to stream the reasoning content in responses, so should we consider making this configurable?