- What can it do?
- Let me give you an example: the use of Responses API
- Add a new feature at a glance:
- Prices and availability
OpenAI announced a major functional extension of Responses API to open up more modular capabilities to developers, making AI applications to construct multi-model, complex task chains easier, more manageable and stronger.
Responses API is used mainly for building intelligent applications, i.e. intelligent applications with the ability to “think for themselves + call for tools”.
** What’s the difference between it and the tradition ChatGPT API?
What can it do?
** Basic function**
- Answer questions, chats, writings, summaries, etc. (like ChatGPT API)
** Add Tool Calling** The following tools can be automatically called by AI:
-
Web search: real-time search for up-to-date information
-
File search: search and analyse files uploaded by users
-
Code Interpreter: running code, drawing, math
-
Image Generation Tool: Images generated from text
-
Remote MCP tool: Connect to other systems (e. g. Stripe, Shopify, Twilio, etc.)
** The enhanced “intelligent thinking”**
-
Models can be “chained thinking” (Chain of Thought), multi-step reasoning and phased-in tools in one response.
-
Support Reasoning Summer: Seeing how models step by step to come up with answers.
** Backstage mode (Background Mode)** AI can be allowed to do complex tasks (such as long search, analysis, drawing) without fear of time-lapse.
Let me give you an example: the use of Responses API
You can build a “financial consultant AI” with Responses API, which can do: Read your Excel report uploaded (searched through file); Analyse the data and generate graphs (by Code Interpreter); Search for recent financial news (through Web search); Finally, a professional financial analysis report is produced. You just call the API once, it’ll automatically phase out all the tasks.
Add a new feature at a glance:
1. ** Remote MCP server support**
-
Developer can now extend context (Context) from local to remote MCP service
-
A tool to connect the model to any MCP server hosted through several lines of code.
-
To build large, multi-source, distributed AI applications
-
Making context construction more modular, safer and shared
-
OpenAI also joined MCP’s steering committee to push the deal forward.
-
MCP suppliers supported include: Cloudflare, Stripe, Twilio, Shopify, PayPal, Zapier, etc.
##2. ** Image Generations Access Responses API**
-
gpt-image-1 model image generation can be used as a tool in collaboration with language models
-
Use the gpt-image-1 model as an in-house tool to support real-time preview and multi-cycle editing.
-
Developers can directly call for image generation in the reasoning process and progressively optimize results.
-
Supports stream preview of images: generating processes that can be visualized in real time
-
In support of multi-wheel image editing: step-by-step refinement of image details to fit the creative scene
#3 Code Interpreter (Advanced Code Tool) Access Responses API
-
Data analysis, image processing, complex mathematical and programming problems can be solved using Responses API.
-
For: Data analysis
-
Mathematical operations.
-
Programming calculations
-
Image understanding task (e.g., “Look at the picture thinking”)
Support OpenAI o3 and o4-mini, upgrade chain reasoning (Chain-of-Thought) • Official statements
##4. ** File search support online**
-
The relevant document paragraph (chunk) can be found using the reasoning model
-
Support: ** Two vector databases searched simultaneously**
-
Array-based Attribute Filtering
View file search examples
##5. New Developer Support
-
Background mode running an asynchronous task without blocking the current request to specify the document
-
Reasoning Summers quick look at the path of reasoning within the model to facilitate debugging and evaluation
-
Encryped Reasoning Items allows cross-requesting re-use of reasoning in ZDR (Zero Data Retion)
Prices and availability
