Structured output.

JSON schema (recommended)

On Chat Completions, pass response_format.type: "json_schema" with your schema. The response content is a single JSON document conforming to it — parse and go.

curl https://api.layerx1.in/v1/chat/completions \
  -H "authorization: Bearer $LAYERX1_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "model": "lx1-gpt-oss-120b",
    "messages": [{ "role": "user", "content": "Extract: Ada Lovelace, born 1815." }],
    "response_format": {
      "type": "json_schema",
      "json_schema": {
        "name": "person",
        "strict": true,
        "schema": {
          "type": "object",
          "properties": {
            "name": { "type": "string" },
            "birth_year": { "type": "integer" }
          },
          "required": ["name", "birth_year"],
          "additionalProperties": false
        }
      }
    }
  }'

{ "name": "Ada Lovelace", "birth_year": 1815 }

JSON mode

When any valid JSON is enough and the exact shape can flex, use { "type": "json_object" } and describe the fields you want in the prompt:

{ "response_format": { "type": "json_object" } }

Guaranteed parseable, not guaranteed shaped — validate before trusting field names.

On the Messages dialect

The Anthropic dialect expresses shapes through tools: define a tool whose input_schema is your output schema and force it with tool_choice — the arguments of the resulting tool_use block are your structured output.

{
  "tools": [{
    "name": "record_person",
    "description": "Record the extracted person",
    "input_schema": {
      "type": "object",
      "properties": {
        "name": { "type": "string" },
        "birth_year": { "type": "integer" }
      },
      "required": ["name", "birth_year"]
    }
  }],
  "tool_choice": { "type": "tool", "name": "record_person" }
}

Practical notes

• Set strict: true and additionalProperties: false — schemas that forbid extras fail loud instead of drifting quietly.
• Keep schemas shallow. Deeply nested optional trees invite empty objects; several flat calls beat one cathedral schema.
• Reasoning models produce excellent structured output but budget max_tokens generously — hidden reasoning counts against it before the JSON is emitted.
• Always validate before acting — a 400 on an impossible constraint or a truncated length finish is recoverable if you check for it (Errors).