prefix_mismatch
OpenAI input_transformations entry reason: "prefix_mismatch".
What it means
Content before this reasoning item was inserted or edited, so GPT-6.2 or later removed the item. The request still succeeded.
On GPT-6.2 and later, you insert or edit content before the reasoning item. For example, rewriting an earlier tool result changes the prefix.
Reasoning models (checked 2026-11-19)
Starting with GPT-6.2, the Responses API drops reasoning items after mid-conversation edits to the conversation history (prefix edits), certain model changes (including downgrades), or changes to the end user or organization.
Latest model (GPT-6.2) (checked 2026-11-19)
Is it yours to fix?
Yes. Your code changed something, so it’s yours to fix.
How to fix it
Don’t insert or edit items before a reasoning item. Append new items at the end of input and replay earlier items unchanged.
An example that produces it (Earlier tool output shortened):
{
"input": [
{
"content": "Run the tests.",
"role": "user"
},
{
"encrypted_content": "gAAAABrs_1",
"id": "rs_1",
"summary": [],
"type": "reasoning"
},
{
"arguments": "{}",
"call_id": "call_1",
"name": "run_tests",
"type": "function_call"
},
{
"call_id": "call_1",
"output": "PASS tests/a.test.js\nPASS tests/b.test.js\nFAIL tests/c.test.js\n Expected 2, received 3\n at line 14",
"type": "function_call_output"
}
],
"model": "gpt-6.2-astra"
}{
"input": [
{
"content": "Run the tests.",
"role": "user"
},
{
"encrypted_content": "gAAAABrs_1",
"id": "rs_1",
"summary": [],
"type": "reasoning"
},
{
"arguments": "{}",
"call_id": "call_1",
"name": "run_tests",
"type": "function_call"
},
{
"call_id": "call_1",
"output": "FAIL tests/c.test.js",
"type": "function_call_output"
},
{
"encrypted_content": "gAAAABrs_2",
"id": "rs_2",
"summary": [],
"type": "reasoning"
},
{
"content": [
{
"text": "One test fails.",
"type": "output_text"
}
],
"role": "assistant",
"type": "message"
},
{
"content": "Fix it.",
"role": "user"
}
],
"model": "gpt-6.2-astra"
}Related reason codes
end_user_mismatch: end_user_mismatch: safety_identifier differs between requestsmodel_mismatch: model_mismatch: reasoning removed after a model switchorganization_mismatch: organization_mismatch: reasoning item from another organizationturn_boundary: turn_boundary: earlier-turn reasoning not used
Last checked 2026-11-19 against the official OpenAI docs.
FAQ
Does deleting old items cause prefix_mismatch?
Not by itself. OpenAI’s reasoning guide says deleting whole items without editing the remaining items, or appending new items after the reasoning item, does not trigger this reason.
How do I change reasoning effort or tools without it?
Append a configuration_update item for effort or format, or an additional_tools item for tools, instead of changing the request-level parameters.