llm-structured-outputslisted

# LLM Structured Outputs Cross-provider reference for schema-constrained JSON decoding with all major frontier LLM APIs. Last updated: February 23, 2026. ## When to Use This Skill - Implementing structured LLM outputs with any major provider - Choosing the right API field for `json_schema` per vendor - Writing Zod schemas and converting them to JSON Schema for API requests - Building provider-agnostic LLM abstraction layers - Debugging parse failures or schema non-compliance - Migrating off `zod-to-json-schema` (deprecated Nov 2025) - Handling edge cases: truncation, refusals, cache invalidation --- ## Core Concept All four providers support **constrained decoding**: you send a JSON Schema and the model is constrained at the token level to produce compliant JSON. This is mathematical, not probabilistic. | Provider | API Field | Type Value | | --------------------------------- | ------------------------------------- | ------------------- | | Anthropic Claude 4.6 | `output_config.format` | `"json_schema"` | | OpenAI GPT-5.3 (Responses API) | `text.format` | `"json_schema"` | | OpenAI GPT-5.3 (Chat Completions) | `response_format` | `"json_schema"` | | Gemini 3.1 Pro | `generationConfig.responseJsonSchema` | N/A (schema object) | | xAI Grok 4 | `response_format` |