Limitations on enum values in Structured Outputs

API
,
1

The documentation for Structured Outputs just says that Enums are supported, but that doesn’t seem to be entirely true. So far I’ve seen the API reject schemas with enum values containing double-quotes and newlines. What is the full set of allowed characters?

2 Likes
2

This is a question that can be answered by merely sending over five-hundred requests (or 100000+ if you want to find out about all of Unicode), building that many more structured schemas server-side for every one-byte enum. (decode(‘latin1’) to map them, which includes bytes that would be invalid utf-8 start bytes).

This coded task also involves the AI producing without error, which is recorded as true. By being strict, the API structured schema allows only that enum token (or token sequence) even if the AI couldn’t understand usage.

# Construct the json_schema with the current test_value as the enum
json_schema = {
    "name": "ascii_test",
    "description": "A basic structured output response schema for an ASCII test with a fixed value",
    "strict": True,
    "schema": {
        "type": "object",
        "required": [
            "test_value"
        ],
        "properties": {
            "test_value": {
                "type": "string",
                "description": "The value under test which constrains the AI to one possible response",
                "enum": [
                    test_value
                ]
            }
        },
        "additionalProperties": False
    }
}

Look up your character!

Summary
Character With Escape Unescaped
\x00 500 500
\x01 true true
\x02 true true
\x03 true true
\x04 true true
\x05 true true
\x06 true true
\x07 true true
\x08 true true
\x09 true true
\x0a 500 400
\x0b true true
\x0c true true
\x0d true true
\x0e true true
\x0f true true
\x10 true true
\x11 true true
\x12 true true
\x13 true true
\x14 true true
\x15 true true
\x16 true true
\x17 true true
\x18 true true
\x19 true true
\x1a true true
\x1b true true
\x1c true true
\x1d true true
\x1e true true
\x1f true true
’ ’ true true
‘!’ true true
‘"’ 400 400
‘#’ true true
‘$’ true true
‘%’ true true
‘&’ true true
‘’’ true true
‘(’ true true
‘)’ true true
‘*’ true true
‘+’ true true
‘,’ true true
‘-’ true true
‘.’ true true
‘/’ true true
‘0’ true true
‘1’ true true
‘2’ true true
‘3’ true true
‘4’ true true
‘5’ true true
‘6’ true true
‘7’ true true
‘8’ true true
‘9’ true true
‘:’ true true
‘;’ true true
‘<’ true true
‘=’ true true
‘>’ true true
‘?’ true true
‘@’ true true
‘A’ true true
‘B’ true true
‘C’ true true
‘D’ true true
‘E’ true true
‘F’ true true
‘G’ true true
‘H’ true true
‘I’ true true
‘J’ true true
‘K’ true true
‘L’ true true
‘M’ true true
‘N’ true true
‘O’ true true
‘P’ true true
‘Q’ true true
‘R’ true true
‘S’ true true
‘T’ true true
‘U’ true true
‘V’ true true
‘W’ true true
‘X’ true true
‘Y’ true true
‘Z’ true true
‘[’ true true
‘\’ true 500
‘]’ true true
‘^’ true true
‘_’ true true
‘`’ true true
‘a’ true true
‘b’ true true
‘c’ true true
‘d’ true true
‘e’ true true
‘f’ true true
‘g’ true true
‘h’ true true
‘i’ true true
‘j’ true true
‘k’ true true
‘l’ true true
‘m’ true true
‘n’ true true
‘o’ true true
‘p’ true true
‘q’ true true
‘r’ true true
‘s’ true true
‘t’ true true
‘u’ true true
‘v’ true true
‘w’ true true
‘x’ true true
‘y’ true true
‘z’ true true
‘{’ true true
’ ’ true
‘}’ true true
‘~’ true true
\x7f true true
\x80 true true
\x81 true true
\x82 true true
\x83 true true
\x84 true true
\x85 true true
\x86 true true
\x87 true true
\x88 true true
\x89 true true
\x8a true true
\x8b true true
\x8c true true
\x8d true true
\x8e true true
\x8f true true
\x90 true true
\x91 true true
\x92 true true
\x93 true true
\x94 true true
\x95 true true
\x96 true true
\x97 true true
\x98 true true
\x99 true true
\x9a true true
\x9b true true
\x9c true true
\x9d true true
\x9e true true
\x9f true true
\xa0 true true
‘¡’ true true
‘¢’ true true
‘£’ true true
‘¤’ true true
‘¥’ true true
‘¦’ true true
‘§’ true true
‘¨’ true true
‘©’ true true
‘ª’ true true
‘«’ true true
‘¬’ true true
\xad true true
‘®’ true true
‘¯’ true true
‘°’ true true
‘±’ true true
‘²’ true true
‘³’ true true
‘´’ true true
‘µ’ true true
‘¶’ true true
‘·’ true true
‘¸’ true true
‘¹’ true true
‘º’ true true
‘»’ true true
‘¼’ true true
‘½’ true true
‘¾’ true true
‘¿’ true true
‘À’ true true
‘Á’ true true
‘Â’ true true
‘Ã’ true true
‘Ä’ true true
‘Å’ true true
‘Æ’ true true
‘Ç’ true true
‘È’ true true
‘É’ true true
‘Ê’ true true
‘Ë’ true true
‘Ì’ true true
‘Í’ true true
‘Î’ true true
‘Ï’ true true
‘Ð’ true true
‘Ñ’ true true
‘Ò’ true true
‘Ó’ true true
‘Ô’ true true
‘Õ’ true true
‘Ö’ true true
‘×’ true true
‘Ø’ true true
‘Ù’ true true
‘Ú’ true true
‘Û’ true true
‘Ü’ true true
‘Ý’ true true
‘Þ’ true true
‘ß’ true true
‘à’ true true
‘á’ true true
‘â’ true true
‘ã’ true true
‘ä’ true true
‘å’ true true
‘æ’ true true
‘ç’ true true
‘è’ true true
‘é’ true true
‘ê’ true true
‘ë’ true true
‘ì’ true true
‘í’ true true
‘î’ true true
‘ï’ true true
‘ð’ true true
‘ñ’ true true
‘ò’ true true
‘ó’ true true
‘ô’ true true
‘õ’ true true
‘ö’ true true
‘÷’ true true
‘ø’ true true
‘ù’ true true
‘ú’ true true
‘û’ true true
‘ü’ true true
‘ý’ true true
‘þ’ true true
‘ÿ’ true true

Yes, even unprintable bytes:

"content": "{\\"test_value\\":\\"\\u0004\\"}"

The 400 errors:

\" is not allowed in string literals for structured outputs (strict=true)
\n is not allowed in string literals for structured outputs (strict=true)