OpenAI Responses API Updates May 2025

OpenAI has introduced several new tools and features to the Responses API to enhance the development of agentic applications. These updates include support for remote Model Context Protocol (MCP) servers, the integration of image generation and Code Interpreter, and new reliability and privacy features such as background mode and encrypted reasoning items.

Remote Model Context Protocol (MCP) Server Support

OpenAI now supports remote MCP servers within the Responses API, allowing developers to connect models to tools hosted on any MCP server with minimal code. MCP is an open protocol designed to standardize how applications provide context to LLMs.

OpenAI has joined the MCP steering committee to support this developing standard. Popular remote MCP servers currently available include:

  • Cloudflare
  • HubSpot
  • Intercom
  • PayPal
  • Plaid
  • Shopify
  • Stripe
  • Square
  • Twilio
  • Zapier

Enhanced Built-in Tools

Developers can now access several high-performance tools through a single API call. Models such as o3 and o4-mini can now call these tools directly within their chain-of-thought, which has improved performance on benchmarks like Humanity’s Last Exam.

Image Generation

The gpt-image-1 model is now available as a tool within the Responses API. This integration supports:

  • Real-time streaming: Developers can view previews of images as they are generated.
  • Multi-turn edits: Models can refine images step-by-step through granular prompting.

Code Interpreter

The Code Interpreter tool is now integrated into the Responses API, facilitating data analysis, complex mathematical problem solving, and image manipulation. For o3 and o4-mini, using Code Interpreter within the chain-of-thought has led to improved benchmark performance.

File Search

File search is now available for reasoning models. This tool allows developers to retrieve relevant document chunks based on user queries. Updates include support for attribute filtering with arrays and the ability to perform searches across multiple vector stores.

Reliability, Visibility, and Privacy Features

OpenAI has introduced three new features to the Responses API to improve the developer and enterprise experience:

  • Background Mode: To handle long-running tasks that may take several minutes (similar to Deep Research or Operator), background mode allows tasks to be executed asynchronously. Developers can poll for completion or stream events to track state.
  • Reasoning Summaries: The API can now generate natural-language summaries of the model's internal chain-of-thought at no additional cost, aiding in debugging and auditing.
  • Encrypted Reasoning Items: For customers eligible for Zero Data Retention (ZDR), reasoning items can be reused across API requests without being stored on OpenAI servers. This increases cache hit rates and reduces costs and latency for o3 and o4-mini.

Model Availability and Pricing

All new tools and features are supported across the GPT-4o series, GPT-4.1 series, and the OpenAI o-series reasoning models (o1, o3, o3-mini, and o4-mini), with the exception of image generation, which is only supported on o3.

Pricing Details

  • Image Generation: $5.00/1M text input tokens, $10.00/1M image input tokens, and $40.00/1M image output tokens (75% discount for cached input tokens).
  • Code Interpreter: $0.03 per container.
  • File Search: $0.10/GB of vector storage per day and $2.50/1k tool calls.
  • Remote MCP Servers: No additional cost to call the tool; users are billed only for API output tokens.

Sources