The OpenAI Responses API is the company's core API primitive for building agentic applications. OpenAI introduced it in March 2025 as part of a broader set of tools for building agents, and expanded it on May 21, 2025 with more built-in tools and enterprise-oriented features.[1][2]
OpenAI presents the Responses API as the successor direction for agent-building on its platform. It combines ordinary text generation with tool use, persistent response objects, and support for OpenAI-hosted capabilities such as web search, file search, and computer use.[1][2]
At launch, OpenAI described the Responses API as a new primitive that combined the simplicity of Chat Completions with the tool-use capabilities that had previously been spread across the Assistants stack. The first release supported built-in tools such as web search, file search, and computer use, and shipped alongside the Agents SDK.[1]
On May 21, 2025 OpenAI added several new capabilities to the API:[2]
| Addition | What OpenAI added |
|---|---|
| Remote MCP support | Ability to connect to remote Model Context Protocol servers |
| Image generation | gpt-image-1 as a Responses API tool |
| Code Interpreter | Hosted containerized code execution |
| File search upgrades | Multi-vector-store search and attribute filtering |
| Background mode | Asynchronous long-running tasks |
| Reasoning summaries | Natural-language summaries of internal reasoning |
| Encrypted reasoning items | Reusable reasoning state for Zero Data Retention users |
OpenAI also said that reasoning models such as o3 and o4-mini could call tools directly within their chain of thought inside the Responses API, improving both usefulness and efficiency.[2]
OpenAI kept the same basic pricing model for existing tools while documenting the per-tool charges in the May 2025 update.[2]
| Tool or feature | Price noted by OpenAI |
|---|---|
| Image generation | $5.00 per 1M text input tokens, $10.00 per 1M image input tokens, $40.00 per 1M image output tokens |
| Code Interpreter | $0.03 per container |
| File search | $0.10 per GB of vector storage per day and $2.50 per 1,000 tool calls |
| Remote MCP tool | No extra tool fee beyond normal output-token billing |
The Responses API mattered because it moved OpenAI's platform toward a single agent-building surface instead of forcing developers to stitch together multiple APIs. By late May 2025 it supported hosted tools, remote MCP servers, asynchronous jobs, and structured reasoning features in one interface.[1][2][3]