Strict Mode

~22 min · strict, schema-validation

Level 0Tokenizer

0 XP0/54 lessons0/10 achievements

0/120 XP to next level120 XP to go0% complete

Setting strict: true guarantees the model's output matches your JSON Schema exactly. This eliminates the need for output validation — the model is constrained to produce conforming output.

Requirements for Strict Mode

additionalProperties must be false on every object in the schema.
All fields in properties must be listed in required.
Optional fields must use a null type union: "type": ["string", "null"].
The model is guaranteed to produce outputs matching the schema.

Strict mode uses constrained decoding at the model level — it's not just validation after the fact. The model literally cannot produce non-conforming output, even for complex nested schemas.

Code

Tool with strict=True·python

{
    "type": "function",
    "name": "create_event",
    "description": "Create a calendar event",
    "parameters": {
        "type": "object",
        "properties": {
            "title": {"type": "string"},
            "date": {"type": "string", "description": "ISO 8601 date"},
            "description": {"type": ["string", "null"]},  # optional via null union
        },
        "required": ["title", "date", "description"],     # ALL fields required
        "additionalProperties": False,                     # mandatory for strict
    },
    "strict": True,
}

External links

Structured outputs (strict)

Exercise

Take a tool that accepts an enum field. Without strict, prompt the model in a way that tempts an invalid enum value. Watch it return one. Now flip strict=True and watch the API reject the bad call before it reaches you.