Summary
After upgrading to OpenClaw 2026.6.1 and running the config/plugin migration, OpenAI/ChatGPT Responses inference fails for openai/gpt-5.4 and openai/gpt-5.5 with:
causeCode=invalid_provider_content_typeapi=openclaw-openai-responses-transportmessage=Connection error
Local vLLM inference succeeds in the same install, and gateway/plugin health is otherwise clean.
Environment
- OpenClaw: 2026.6.1
2e08f0f
- OS: Linux
7.0.10-arch1-1 x64
- Node: 22.22.2
- Install/update method: npm
- Gateway: systemd user service, running/reachable
- Auth: OpenAI OAuth profile reports OK
- Default model:
openai/gpt-5.5
- Session context shown by status:
gpt-5.5 (1000k ctx)
Config shape
The migrated config uses:
{
"models.providers.openai": {
"api": "openai-chatgpt-responses",
"auth": "oauth",
"baseUrl": "https://chatgpt.com/backend-api"
},
"agents.defaults.model": {
"primary": "openai/gpt-5.5"
}
}Configured OpenAI models include:
gpt-5.4, api openai-chatgpt-responses, runtime pigpt-5.5, api openai-chatgpt-responses, runtime pigpt-5.3-codex, api openai-chatgpt-responses
Reproduction
openclaw infer model run --model openai/gpt-5.5 --prompt "reply OK"
openclaw infer model run --model openai/gpt-5.4 --prompt "reply OK"
openclaw infer model run --model vllm/gemma4-31b --prompt "reply OK"
Actual result
openai/gpt-5.5:
[openai-transport] [responses] error provider=openai api=openclaw-openai-responses-transport model=gpt-5.5 name=Error status=undefined code=undefined type=undefined causeName=ProviderHttpError causeCode=invalid_provider_content_type message=Connection error.
Error: No text output returned for provider "openai" model "gpt-5.5": Connection error..
openai/gpt-5.4:
[openai-transport] [responses] error provider=openai api=openclaw-openai-responses-transport model=gpt-5.4 name=Error status=undefined code=undefined type=undefined causeName=ProviderHttpError causeCode=invalid_provider_content_type message=Connection error.
Error: No text output returned for provider "openai" model "gpt-5.4": Connection error..
openai/gpt-5.3-codex:
[openai-transport] [responses] error provider=openai api=openclaw-openai-responses-transport model=gpt-5.3-codex name=Error status=400 code=undefined type=undefined causeName=undefined causeCode=undefined message=400 status code (no body)
Local vLLM succeeds:
model.run via local
provider: vllm
model: gemma4-31b
outputs: 1
OK
Expected result
openai/gpt-5.4 and openai/gpt-5.5 should return text output, or at least surface the underlying HTTP response/status/body/content-type clearly enough to diagnose.
Notes
This appears distinct from the plugin migration failure mode. Doctor shows plugin registry clean:
- Plugins loaded: 71
- Plugin errors: 0
- no
runtime.llm.complete
- no conflicting legacy plugin index warning after cleanup
This looks like a 2026.6.1 OpenAI/ChatGPT Responses transport issue where the transport expects SSE/JSON but receives another content type from the backend.
A redacted repro bundle is available if useful, but I am not attaching it publicly because it contains local status/log context.
Possibly related historical/current issues:
Summary
After upgrading to OpenClaw 2026.6.1 and running the config/plugin migration, OpenAI/ChatGPT Responses inference fails for
openai/gpt-5.4andopenai/gpt-5.5with:causeCode=invalid_provider_content_typeapi=openclaw-openai-responses-transportmessage=Connection errorLocal vLLM inference succeeds in the same install, and gateway/plugin health is otherwise clean.
Environment
2e08f0f7.0.10-arch1-1x64openai/gpt-5.5gpt-5.5 (1000k ctx)Config shape
The migrated config uses:
{ "models.providers.openai": { "api": "openai-chatgpt-responses", "auth": "oauth", "baseUrl": "https://chatgpt.com/backend-api" }, "agents.defaults.model": { "primary": "openai/gpt-5.5" } }Configured OpenAI models include:
gpt-5.4, apiopenai-chatgpt-responses, runtimepigpt-5.5, apiopenai-chatgpt-responses, runtimepigpt-5.3-codex, apiopenai-chatgpt-responsesReproduction
Actual result
openai/gpt-5.5:openai/gpt-5.4:openai/gpt-5.3-codex:Local vLLM succeeds:
Expected result
openai/gpt-5.4andopenai/gpt-5.5should return text output, or at least surface the underlying HTTP response/status/body/content-type clearly enough to diagnose.Notes
This appears distinct from the plugin migration failure mode. Doctor shows plugin registry clean:
runtime.llm.completeThis looks like a 2026.6.1 OpenAI/ChatGPT Responses transport issue where the transport expects SSE/JSON but receives another content type from the backend.
A redacted repro bundle is available if useful, but I am not attaching it publicly because it contains local status/log context.
Possibly related historical/current issues: