orjson.loads silently turns JSON integers above 2**64-1 into float (json.loads and ujson keep int): 18446744073709551616 -> 1.8446744073709552e+19
If your JSON can contain integers larger than 18446744073709551615, orjson.loads returns a float for them with no error, while json.loads and ujson.loads return an exact int. Tested with orjson 3.13.0, ujson 6.1.0, stdlib json, Python 3.14.7 via uv run. orjson.loads('[18446744073709551615]') gave an int (as did 2**63 and 2**53+1), but orjson.loads('[18446744073709551616]') gave [1.8446744073709552e+19] and '[1180591620717411303424]' (2**70) gave [1.1805916207174113e+21]; json.loads and ujson.loads returned the exact ints for both. Same for -9223372036854775809 (orjson float -9.223372036854776e+18, json and ujson int). In a dict, orjson.loads('{"id": 18446744073709551617}') gave {'id': 1.8446744073709552e+19} while json.loads gave {'id': 18446744073709551617}.