Is serialisation deterministic? #12067

I would distinguish repeatable in a particular case from canonical serialization.

For ordinary models containing only ordered values, Pydantic preserves model-field order and Python dictionaries preserve insertion order, so repeated model_dump() / model_dump_json() calls will normally produce the same result.

But Pydantic does not turn arbitrary data into a canonical representation. Output can vary when the value graph contains, for example:

• sets or frozensets;
• dictionaries constructed in different insertion orders;
• custom serializers;
• values whose representation depends on configuration or context;
• non-finite floats or other types whose JSON representation is implementation-dependent.

model_dump() is also a Python-object representation, so equality of two dumps does not imply stable byte encoding.

If the output will be hashed, signed, cached by bytes, or used as an idempotency key, normalize unordered values and perform an explicit canonical JSON step:

import json

data = model.model_dump(mode="json")

canonical = json.dumps(
    data,
    sort_keys=True,
    separators=(",", ":"),
    ensure_ascii=False,
    allow_nan=False,
).encode("utf-8")

For a set-like field, make its ordering explicit before that step, preferably in a serializer:

from pydantic import BaseModel, field_serializer


class Example(BaseModel):
    tags: set[str]

    @field_serializer("tags")
    def serialize_tags(self, tags: set[str]) -> list[str]:
        return sorted(tags)

Also note that sort_keys=True is deterministic JSON, but it is not by itself a cross-language canonicalization standard. If signatures must be verified by other implementations, use a defined scheme such as RFC 8785 (JCS) and test the exact byte vectors.

So: deterministic for common ordered inputs in the same configuration, yes; a blanket canonical-output guarantee for every supported value, no.

Reference: https://docs.pydantic.dev/latest/concepts/serialization/

If this solves your issue, please consider marking this comment as the answer!