Initial Checks

• I'm using the latest version of Pydantic AI
• I've searched for my issue in the issue tracker before opening this issue

Description

After upgrading pydantic-ai-slim from 1.63.0 to 1.64.0+, the behavior of result.new_messages() appears to have changed.

When calling agent.run() with message_history (and without user_prompt), the messages supplied in message_history used to be treated purely as prior context, so result.new_messages() only returned messages generated during the run.

Starting in 1.64.0, new_messages() also returns the user prompt that was provided inside message_history.

For example, the following code now returns the original ModelRequest as part of new_messages() even though it was part of the input history.

This seems related to changes introduced in this PR:
https://github.com/pydantic/pydantic-ai/pull/4419/changes#diff-71730070455237accebd709ba7fe41ce60edbd238af02e7d86a208cf8f10ab12R1543-R1545

The last version that behaves as expected for me is 1.63.0.

Is this an intended change, or a regression? If intentional, is there a recommended way to get only the messages generated during the run?

Old behavior (≤1.63.0)

new_messages() returns only messages generated during the run.

New behavior (≥1.64.0)

new_messages() includes the ModelRequest from message_history, meaning the user prompt "hi" is returned even though it was already part of the provided history.

Minimal, Reproducible Example

result = await agent.run(
    message_history=[
        ModelRequest(
            parts=[
                UserPromptPart(
                    content="hi"
                )
            ]
        )
    ],
    deps=deps,
)

print(result.new_messages())

Logfire Trace

No response

Python, Pydantic AI & LLM client version

• Python: 3.12
• Pydantic AI: 1.68.0
• LLM provider SDK: doesn't matter