Home / OpenAI API pricing / OpenAI Responses API
OpenAI Responses API: what it is and what it costs
The Responses API is OpenAI’s newer way to call its models, and the one it recommends for new projects. It has no price of its own: you pay the model’s token prices, plus fees for any built-in tools the model uses, such as web search.
Checked October 20, 2026 against OpenAI’s migration guide and pricing page · token prices checked October 13, 2026
store: falseWhat the Responses API is
OpenAI calls the Responses API its new “API primitive”, an evolution of Chat Completions. You send a request to /v1/responses and get back a typed list of output items, which can include the model’s text, its reasoning and its tool calls. In one request the model can call several tools, including OpenAI’s own built-in ones:
- web search and file search
- code interpreter and a hosted shell, which run in containers
- computer use, image generation and remote MCP servers
- your own functions, as in Chat Completions
It can also keep a conversation going for you: pass previous_response_id to continue from an earlier response, or use the Conversations API for a conversation that persists. With Chat Completions you resend the whole transcript yourself.
Responses vs Chat Completions
| Chat Completions | Responses | |
|---|---|---|
| Endpoint | /v1/chat/completions | /v1/responses |
| Input and output | An array of messages in, choices out | input (text or items) in, a typed output array out |
| Conversation state | You send the full history each time | previous_response_id, the Conversations API, or your own history |
| OpenAI’s built-in tools | Not available: you write your own | Web search, file search, code interpreter, computer use, image generation, MCP |
| Tools on the newest models | GPT-6 Astra and GPT-6.1 Sol need Responses for tool calling | Supported |
| Structured Outputs | response_format | text.format |
| Several answers per request | Yes, with n | No: one answer per request |
| Stored by default | For new accounts | Yes |
| Status | Still supported | Recommended for all new projects |
What you pay for
OpenAI’s pricing page says it plainly: the Responses API, Chat Completions, the Batch API and the others aren’t priced separately. A request costs:
- Tokens at the model’s prices. Input, cached input and output, the same as in any other OpenAI API.
- Built-in tool fees, when the model uses those tools (table below).
- The conversation so far. With
previous_response_id, every earlier input token in the chain is billed again as input on the next turn. OpenAI lists “assumingprevious_response_idremoves billing for prior context” as a common mistake.
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6 Astra | $10.00 | $1.00 | $50.00 |
| GPT-6.1 Sol | $2.00 | $0.10 | $10.00 |
| GPT-6 Luna | $0.10 | $0.01 | $0.50 |
| Tool | Fee |
|---|---|
| Web search | $10.00 per 1,000 calls, plus the search content tokens at the model’s rates |
| File search | $2.50 per 1,000 calls, plus storage at $0.10 per GB a day after the first 1 GB |
| Code interpreter and hosted shell | Per container, per 20-minute session: 1 GB $0.03, 4 GB $0.12, 16 GB $0.48, 64 GB $1.92; billed by the minute, with a 5-minute minimum |
From OpenAI’s pricing page. Token prices checked October 13, 2026. File search call fees apply to the Responses API only. Tokens the tools use are billed at the model’s rates.
Caching: why Responses can cost less
Prompt caching is on by default. When a request starts with the same text as an earlier one (at least 1,024 tokens on GPT-5.6 and later models), OpenAI reuses its work and bills that part at the cached input price, $0.10 per million tokens on GPT-6.1 Sol instead of $2.00. In its internal tests, OpenAI says, the Responses API used the cache 40% to 80% better than Chat Completions, which lowers costs.
On GPT-5.6 and later models, writing to the cache costs 1.25 times the input price for the default 30 minutes ($2.50 per million on GPT-6.1 Sol). In the Responses API you can ask for a 12-hour cache instead, at 2.5 times the input price ($5.00). OpenAI’s caching guide works out that on a model that reads the cache at a tenth of the input price, writing a prompt prefix once and reading it nine more times costs 2.15 times its normal input price, against 10 times with no caching.
Worked examples
previous_response_id: 1,000-token instructions, 100-token questions, 400-token answers. 33,500 input tokens are billed in all, because each turn reads the whole conversation so far.| Turn | Input tokens billed | Cost of the turn |
|---|---|---|
| 1 | 1,100 | $0.0062 |
| 2 | 1,600 | $0.0072 |
| 5 | 3,100 | $0.0102 |
| 10 | 5,600 | $0.0152 |
Our arithmetic at list prices, checked October 13, 2026. In practice the repeated part of each turn is often read from the cache at the cheaper rate. Try your own numbers in the cost calculator.
Should you move from Chat Completions?
Not in a hurry: OpenAI says Chat Completions remains supported, and suggests moving one feature at a time, starting with plain text generation. Moving makes sense when you want OpenAI’s built-in tools, tool calling on GPT-6 Astra or GPT-6.1 Sol, or the reasoning and caching gains OpenAI reports. The old Assistants API is already gone: OpenAI shut it down on August 26, 2026 and points its users to Responses.
If you might switch providers, check which interface each one copies. xAI and DeepSeek list support for a Responses API, while Anthropic’s and Google’s OpenAI-compatible endpoints are documented for Chat Completions. See OpenAI API alternatives.
Questions
What is the OpenAI Responses API?
OpenAI’s current interface for generating text and running agent-like tasks with its models. It takes text or a list of items, returns typed output items, can keep conversation state, and lets the model use OpenAI’s built-in tools in the same request.
Is the Responses API more expensive than Chat Completions?
No. Neither is priced separately; you pay the same token prices for the model you choose. The Responses API can cost less thanks to better cache use, and more if you add built-in tools, which carry their own fees.
Is Chat Completions being deprecated?
Not as of October 20, 2026: OpenAI’s migration guide says Chat Completions remains supported, while recommending Responses for all new projects.
What happened to the Assistants API?
OpenAI shut it down on August 26, 2026 and it is no longer available. OpenAI’s migration guide moves Assistants integrations to the Responses API.
Does previous_response_id save money?
It saves you resending the conversation yourself, not the tokens: every earlier input token in the chain is still billed as input. Caching is what lowers the cost of the repeated part.
Can I stop OpenAI storing my responses?
Yes. Set store: false. Responses are otherwise kept for 30 days, and you can view them in the dashboard logs; objects in the Conversations API don’t expire after 30 days. For Zero Data Retention organizations, OpenAI applies store: false automatically.
Sources
- OpenAI: Migrate to the Responses API (what it is, differences, Assistants API)
- OpenAI: API pricing (tool fees, APIs not priced separately) and token prices, checked October 13, 2026
- OpenAI: Conversation state (storage, billing of earlier turns) and Prompt caching (cache rates and lifetimes)
- All read October 20, 2026. Copies are kept with our price records.