Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More
OpenAI is rolling out a set of significant updates to its newish Responses API, aiming to make it easier for developers and enterprises to build intelligent, action-oriented agentic applications.
These enhancements include support for remote Model Context Protocol (MCP) servers, integration of image generation and Code Interpreter tools, and upgrades to file search capabilities—all available as of today, May 21.
First launched in March 2025, the Responses API serves as OpenAI’s toolbox for third-party developers to build agentic applications atop some of the core functionalities of its hit services ChatGPT and its first-party AI agents Deep Research and Operator.
In the months since its debut, it has processed trillions of tokens and supported a broad range of use cases, from market research and education to software development and financial analysis.
Popular applications built with the API include Zencoder’s coding agent, Revi’s market intelligence assistant, and MagicSchool’s educational platform.
The basis and purpose of the Responses API
The Responses API debuted alongside OpenAI’s open-source Agents SDK in March 2025, as part of an initiative to provide third-party developer access to the same technologies powering OpenAI’s own AI agents like Deep Research and Operator.
This way, startups and companies outside of OpenAI could integrate the same tech as it offers through ChatGPT into their own products and services, be they internal for employee usage or external for customers and partners.
Initially, the API combined elements from Chat Completions and the Assistants API—delivering built-in tools for web and file search, as well as computer use—enabling developers to build autonomous workflows without complex orchestration logic. OpenAI said at that time that the Chat Completions API would be deprecated by mid 2026.
The Responses API provides visibility into model decisions, access to real-time data, and integration capabilities that allowed agents to retrieve, reason over, and act on information.
This launch marked a shift toward giving developers a unified toolkit for creating production-ready, domain-specific AI agents with minimal friction.
Remote MCP server support broadens integration potential
A key addition in this update is support for remote MCP servers. Developers can now connect OpenAI’s models to external tools and services such as Stripe, Shopify, and Twilio using only a few lines of code. This capability enables the creation of agents that can take actions and interact with systems users already depend on. To support this evolving ecosystem, OpenAI has joined the MCP steering committee.
The update brings new built-in tools to the Responses API that enhance what agents can do within a single API call.
A variant of OpenAI’s hit GPT-4o native image generation model — which inspired a wave of “Studio Ghibli” style anime memes around the web and buckled OpenAI’s servers with its popularity, but can obviously create many other image styles — is now available through the API under the model name “gpt-image-1.” It includes potentially helpful and fairly impressive new features like real-time streaming previews and multi-turn refinement.
This enables developers to build applications that can produce and edit images dynamically in response to user input.
Additionally, the Code Interpreter tool is now integrated into the Responses API, allowing models to handle data analysis, complex math, and logic-based tasks within their reasoning processes.
The tool helps improve model performance across various technical benchmarks and allows for more sophisticated agent behavior.
Improved file search and context handling
The file search functionality has also been upgraded. Developers can now perform searches across multiple vector stores and apply attribute-based filtering to retrieve only the most relevant content.
This improves the precision of information agents use, enhancing their ability to answer complex questions and operate within large knowledge domains.
New enterprises reliability, transparency features
Several features are designed specifically to meet enterprise needs. Background mode allows for long-running asynchronous tasks, addressing issues of timeouts or network interruptions during intensive reasoning.
Reasoning summaries, a new addition, offer natural-language explanations of the model’s internal thought process, helping with debugging and transparency.
Encrypted reasoning items provide an additional privacy layer for Zero Data Retention customers.
These allow models to reuse previous reasoning steps without storing any data on OpenAI servers, improving both security and efficiency.
The latest capabilities are supported across OpenAI’s GPT-4o series, GPT-4.1 series, and the o-series models, including o3 and o4-mini. These models now maintain reasoning state across multiple tool calls and requests, which leads to more accurate responses at lower cost and latency.
Yesterday’s price IS today’s price!
Despite the expanded feature set, OpenAI has confirmed that pricing for the new tools and capabilities within the Responses API will remain consistent with existing rates.
For example, the Code Interpreter tool is priced at $0.03 per session, and file search usage is billed at $2.50 per 1,000 calls, with storage costs of $0.10 per GB per day after the first free gigabyte.
Web search pricing varies based on the model and search context size, ranging from $25 to $50 per 1,000 calls. Image generation through the gpt-image-1 tool is also charged according to resolution and quality tier, starting at $0.011 per image.
All tool usage is billed at the chosen model’s per-token rates, with no additional markup for the newly added capabilities.
What’s next for the Responses API?
With these updates, OpenAI continues to expand what is possible with the Responses API. Developers gain access to a richer set of tools and enterprise-ready features, while enterprises can now build more integrated, capable, and secure AI-driven applications.
All features are live as of May 21, with pricing and implementation details available through OpenAI’s documentation.
