openai/openai-python · error · ValueError

Cannot provide both response_id/starting_after can't be prov

Error message

Cannot provide both response_id/starting_after can't be provided together with ", ".join(new_response_args_names)

What it means

Raised by Responses.stream() when you mix continuation arguments (response_id or starting_after) with arguments that create a brand-new response (model, input, instructions, tools, etc.). The stream API supports two modes — resuming/forking an existing response or creating a new one — and supplying both is ambiguous, so the SDK raises ValueError listing the offending new-response args.

Source

Thrown at src/openai/resources/responses/responses.py:1215

            "prompt_cache_retention": prompt_cache_retention,
            "reasoning": reasoning,
            "safety_identifier": safety_identifier,
            "service_tier": service_tier,
            "store": store,
            "stream_options": stream_options,
            "temperature": temperature,
            "text": text,
            "tool_choice": tool_choice,
            "top_logprobs": top_logprobs,
            "top_p": top_p,
            "truncation": truncation,
            "user": user,
            "background": background,
        }
        new_response_args_names = [k for k, v in new_response_args.items() if is_given(v)]

        if (is_given(response_id) or is_given(starting_after)) and len(new_response_args_names) > 0:
            raise ValueError(
                "Cannot provide both response_id/starting_after can't be provided together with "
                + ", ".join(new_response_args_names)
            )
        tools = _make_tools(tools)
        if len(new_response_args_names) > 0:
            if not is_given(input):
                raise ValueError("input must be provided when creating a new response")

            if not is_given(model):
                raise ValueError("model must be provided when creating a new response")

            if is_given(text_format):
                if not text:
                    text = {}

                if "format" in text:
                    raise TypeError("Cannot mix and match text.format with text_format")

View on GitHub (pinned to 9917c6e28e)

Solutions

1. To continue/fork an existing response: remove model/input/etc. and pass only response_id (plus starting_after/output items as needed)
2. To create a new streamed response: remove response_id and starting_after
3. Audit the call site for leftover parameters from the other mode; the error message names the conflicting arguments

Example fix

# before
stream = client.responses.stream(
    response_id="resp_123", model="gpt-4o", input="hi"
)
# after (continue existing)
stream = client.responses.stream(response_id="resp_123")

NEW_RESPONSE_ARGS = {"model", "input", "instructions", "tools", "temperature", "top_p", "max_output_tokens", "metadata", "text", "reasoning", "user", "background"}
continuing = response_id is not None or starting_after is not None
creating = any(kwargs.get(k) is not None for k in NEW_RESPONSE_ARGS)
if continuing and creating:
    raise ValueError("pick one mode: response_id/starting_after OR new-response args")

def is_pure_continuation(response_id, starting_after, **new_args) -> bool:
    given_new = [k for k, v in new_args.items() if v is not None]
    return (response_id is not None or starting_after is not None) and not given_new

try:
    stream = client.responses.stream(**params)
except ValueError as e:
    if "Cannot provide both" in str(e):
        params.pop("response_id", None); params.pop("starting_after", None)
        stream = client.responses.stream(**params)
    else:
        raise

When it happens

Trigger: client.responses.stream(response_id="resp_...", model="gpt-4o", input=[...]) — any of model/input/instructions/tools/temperature/... given together with response_id or starting_after.

Common situations: Refactoring code from create-then-stream to resume-style streaming while leaving old parameters in place; copy-pasting a full parameter set and adding response_id for retries.

Related errors

• input must be provided when creating a new response
• Didn't receive a `response.completed` event.
• Expected to have received `response.created` before `{event.
• Expected a non-empty value for `response_id` but received {r
• WebSocket connection closed with unsent messages

AI-assisted analysis of openai/openai-python@9917c6e28e (2026-08-28). Data as JSON: /api/errors/24b345bb2defa7d9. Report an issue: GitHub.