OpenAIChatTarget and LiteLLMChatTarget leave extra_body_parameters out of their identifier, so targets that send different requests get the same identity. OpenAIResponseTarget already includes it.
The extra body is merged into every request and can set things like reasoning_effort, tools, or the audio voice (audio_response_config is merged into it), and it even overrides the named params. Repro:
OpenAIChatTarget same identifier hash: True (reasoning_effort low vs high)
OpenAIResponseTarget same identifier hash: False
chat temperature in identifier: 0.0 | sent: 1.5 (temperature=0.0, extra_body {"temperature": 1.5})
chat audio voices same hash: True (alloy vs echo)
litellm same hash: True
So stored results, the shared target identity row, and anything keyed on the target hash can't tell these configs apart, and the stored params can be wrong about what was actually sent.
Fix: add extra_body_parameters to both identifiers, same as the Responses target.
OpenAIChatTarget and LiteLLMChatTarget leave
extra_body_parametersout of their identifier, so targets that send different requests get the same identity. OpenAIResponseTarget already includes it.The extra body is merged into every request and can set things like
reasoning_effort, tools, or the audio voice (audio_response_configis merged into it), and it even overrides the named params. Repro:So stored results, the shared target identity row, and anything keyed on the target hash can't tell these configs apart, and the stored params can be wrong about what was actually sent.
Fix: add
extra_body_parametersto both identifiers, same as the Responses target.