
GPT-5 with support for structured outputs, web search and custom tools
Frequently asked questions
What is gpt-5-structured and what makes it different from regular GPT-5?
gpt-5-structured is OpenAI's GPT-5 deployed via Replicate with native enforcement of structured JSON outputs. Unlike standard GPT-5, it guarantees responses conform to your defined JSON schema every time, eliminating post-processing guesswork. It also adds web search integration, custom tool support, and adjustable reasoning effort for developer-focused workflows.
Does gpt-5-structured support real-time web search?
Yes. gpt-5-structured includes a web search integration that allows the model to retrieve current, factual information beyond its training cutoff. This grounds responses in up-to-date data, making it suitable for tasks requiring recent news, live pricing, or current events alongside structured output generation.
What JSON schema features does gpt-5-structured support?
gpt-5-structured enforces strict JSON schema conformance, supporting typed data structures, nested objects, and custom schema definitions. Every response is guaranteed to match your specified schema, removing the need for manual parsing or error handling. This makes it ideal for API integrations, database pipelines, and automated content workflows.
What are the three GPT-5 model tiers available in gpt-5-structured?
gpt-5-structured offers three tiers: gpt-5 (full capability for complex tasks), gpt-5-mini (balanced speed and intelligence for mid-range tasks), and gpt-5-nano (fastest and most cost-efficient for lightweight tasks). Developers can choose the tier that best matches their performance, latency, and budget requirements.
Can gpt-5-structured handle image inputs?
Yes. gpt-5-structured supports multimodal image input, allowing you to pass lists of images alongside text prompts. The model can reason over visual content for tasks like image analysis, content extraction, or creative generation, all while still returning responses in your defined structured format.
How does custom tool support work in gpt-5-structured?
gpt-5-structured supports function calling, letting developers define and pass custom tools that the model can invoke during a conversation. These tools can connect to external APIs, databases, or services, effectively extending the model's capabilities beyond its built-in knowledge and enabling dynamic, real-world integrations within structured workflows.
How do I control response length and reasoning depth in gpt-5-structured?
gpt-5-structured offers two control levers: reasoning effort (minimal, medium, high) adjusts how deeply the model thinks before responding, trading speed for thoroughness; verbosity (low, medium, high) controls output length from concise summaries to comprehensive analyses. Together they let developers fine-tune quality, speed, and cost per request.
Does gpt-5-structured support multi-turn conversations?
Yes. gpt-5-structured supports multi-turn conversations using the previous_response_id parameter. This maintains conversational context across multiple requests without requiring you to resend the full conversation history each time, reducing token usage and simplifying stateful application development.


