Gemini Structured Output Profile

Table of Contents

• Source
• Documented Requirements
  • Supported schema keyword groups
  • Nullability
  • Model behavior note for property order
  • Language vs model differences
• Profile Files
• Requirement Reference Enums
• Notes
• Validation That Must Run In Code

---

This page captures the JSON Schema subset documented by Google for Gemini structured output.

Source

• Google Gemini docs: Structured output - JSON Schema support
• Profile baseline version: 202602 (February 2026)
• Upstream page update shown by Google: February 26, 2026 (UTC)

Documented Requirements

Supported schema keyword groups

• Type values: string, integer, number, boolean, array, object, null
• Generic fields: title, description
• string: enum, format
• integer and number: format, minimum, maximum, enum
• array: minItems, maxItems, items, prefixItems
• object: properties, required, propertyOrdering, additionalProperties

Nullability

• A nullable field uses a type array containing null, for example type: ["string", "null"].

Model behavior note for property order

• The docs state that Gemini API and SDK preserve schema property ordering.
• For Gemini 2.0 and earlier, examples and schemas should include propertyOrdering[].

Language vs model differences

• The documented JSON Schema subset is API-level and is not different by SDK language.
• SDKs can provide different helpers (for example class/type wrappers), but the transmitted schema restrictions are the same.
• The documented difference is model-generation behavior (not SDK language), specifically the Gemini 2.0 propertyOrdering requirement note above.

Profile Files

• gemini_202602.yaml: baseline Gemini documented subset.
• gemini_202503.yaml: stricter Gemini 2.0 profile that requires propertyOrdering on object schemas.

Requirement Reference Enums

• GEMINI_202602
• GEMINI_202503 - older version for Gemini 2.0 models

Notes

• The profile schema encodes only the subset explicitly documented on the linked page.
• This profile intentionally stays conservative where the page does not explicitly document additional JSON Schema keywords.

Validation That Must Run In Code

Even when a schema passes gemini_202602.yaml, additional checks are still useful in application code:

• If targeting Gemini 2.0 or earlier, enforce project policy that object schemas include propertyOrdering.
• When propertyOrdering is present, verify every name exists in properties and that no duplicate names are present.
• Validate semantic consistency between required and properties (every required field is declared as a property).
• Apply model-target checks at runtime because support details can differ by model generation even within the documented keyword subset.

For Gemini 2.0 targets, prefer validating against gemini_202503.yaml first, then run model-target checks in code.