Structured Outputs
A structured output capability introduced by OpenAI in 2024. It constructs a finite state machine (FSM) at each generation step to mask tokens that would violate the target schema, enforcing schema-compliant output at the token sampling level.
Core Facts
OpenAI 的 Function Calling 和 Anthropic 的 Tool Use API 都是 ReAct 模式的标准化实现
JSON Mode only guarantees that the output is valid JSON without constraining the field structure; Structured Outputs enforces output conforming to a predefined JSON Schema at the syntactic level by applying a finite state machine (FSM) to constrain the token sampling path during the decoding phase.
In OpenAI's Function Calling, developers declare function names, parameter types, and descriptions in advance, and the LLM autonomously determines when and which function to call during inference, returning a structured call request to the application layer for execution.
Timeline (last 90 days)
When modifying the Schema of Function Calling or Structured Outputs, active conversations do not automatically reset, and old conversation turns and replayed tool calls may emit the old data shape, causing subsequent conversation turns to crash.
OpenAI's Structured Outputs require additionalProperties: false in the JSON Schema and that all fields be listed in required, with optional fields expressed as nullable.
OpenAI's official guidance draws three lines: the Decisions API for classification, routing, scoring, and gating that require a determination; Structured Outputs (Responses API) for scenarios that require extracting fields or written explanations in a structured form; and function calling for scenarios where the model triggers tools with parameters.
Structured output is a key capability of LLMs in enterprise process integration, achievable by constraining output format through system prompts or by calling OpenAI's Function Calling / Structured Outputs interface.
The structured output design philosophy of the Jev API is similar in goal to OpenAI's Structured Outputs and Anthropic's Tool Use.
OpenAI launched a product called the Decisions API.
OpenAI's structured output feature requires that all object fields list required, disallows additionalProperties, and has specific restrictions on the use of allOf.
OpenAI API historical updates include new model versions (such as the gpt-4o series), expanded context windows, support for Structured Outputs, and the launch of the Batch API.
OpenAI's Structured Outputs mode enforces JSON Schema-compliant output at the token sampling level through constrained decoding technology, completely eliminating formatting errors, a hard constraint that ordinary Function Calling does not possess.
All Facts (16)
OpenAI 的 Function Calling 和 Anthropic 的 Tool Use API 都是 ReAct 模式的标准化实现
80%VerifiedJSON Mode only guarantees that the output is valid JSON without constraining the field structure; Structured Outputs enforces output conforming to a predefined JSON Schema at the syntactic level by applying a finite state machine (FSM) to constrain the token sampling path during the decoding phase.
65%VerifiedIn OpenAI's Function Calling, developers declare function names, parameter types, and descriptions in advance, and the LLM autonomously determines when and which function to call during inference, returning a structured call request to the application layer for execution.
65%VerifiedConstrained decoding restricts the model, via a predefined JSON Schema, so that each generated token can only be selected from the set of valid tokens.
65%VerifiedOpenAI launched a product called the Decisions API.
75%VerifiedOpenAI launched the Structured Outputs feature in August 2024, whose underlying implementation is grammar-constrained sampling technology.
75%Unverified启用strict模式后，模型的输出保证100%符合JSON Schema定义
80%UnverifiedStructured Outputs通过约束解码（Constrained Decoding）技术实现，在模型生成token时动态限制可选token的范围
80%UnverifiedWhen modifying the Schema of Function Calling or Structured Outputs, active conversations do not automatically reset, and old conversation turns and replayed tool calls may emit the old data shape, causing subsequent conversation turns to crash.
50%UnverifiedOpenAI's Structured Outputs require additionalProperties: false in the JSON Schema and that all fields be listed in required, with optional fields expressed as nullable.
50%UnverifiedOpenAI's official guidance draws three lines: the Decisions API for classification, routing, scoring, and gating that require a determination; Structured Outputs (Responses API) for scenarios that require extracting fields or written explanations in a structured form; and function calling for scenarios where the model triggers tools with parameters.
50%UnverifiedStructured output is a key capability of LLMs in enterprise process integration, achievable by constraining output format through system prompts or by calling OpenAI's Function Calling / Structured Outputs interface.
50%UnverifiedThe structured output design philosophy of the Jev API is similar in goal to OpenAI's Structured Outputs and Anthropic's Tool Use.
50%UnverifiedOpenAI's structured output feature requires that all object fields list required, disallows additionalProperties, and has specific restrictions on the use of allOf.
50%UnverifiedOpenAI API historical updates include new model versions (such as the gpt-4o series), expanded context windows, support for Structured Outputs, and the launch of the Batch API.
50%UnverifiedOpenAI's Structured Outputs mode enforces JSON Schema-compliant output at the token sampling level through constrained decoding technology, completely eliminating formatting errors, a hard constraint that ordinary Function Calling does not possess.
50%