OpenAI's Responses API Now Pulls Real Web Images Into AI Apps

OpenAI's Responses API web search tool now returns image results alongside text, enabling apps to surface product photos, places, and visual references with source links.

·
·
·
Read5 min
TypeNews
TopicApi · Image
  • OpenAI's Responses API web search tool now returns image results in addition to text results.
  • Each image result includes a source URL, satisfying OpenAI's requirement that citations be visible and clickable in your UI.
  • Target use cases include product discovery, travel apps, design mood boards, and visual research assistants.
  • No API changes needed if you already use { "type": "web_search" }; the feature is available now.
  • Pricing follows the existing web search model: $2.50 per 1,000 tool calls, plus standard per-token rates for search content.
  • Domain filtering (up to 100 allowed or blocked domains) applies to image results, giving you control over content sources.

OpenAI just expanded what its Responses API web search tool can return. Until now, web search in the Responses API only surfaced text results. With this update, it can also return image results, complete with source URLs, giving your app a way to pull in real-world visuals from the live web on demand.

What changed, exactly

Web search allows models to access up-to-date information from the internet and provide answers with sourced citations. The new addition extends that same pipeline to images. When a user query calls for visual content, the tool can now return image results alongside the usual text, and each result comes with a source link so you can display proper attribution.

The feature is part of the broader web_search tool in the Responses API. For new Responses API integrations, use { "type": "web_search" }. The earlier web_search_preview tool remains available for legacy integrations, but it does not support newer controls such as filters, external_web_access, and return_token_budget.

The problem it solves

Before this, if you wanted to surface images in an AI-powered app, you had two bad options: generate images from scratch using a model like GPT Image, or build a separate integration with a third-party image search API and stitch the results together manually. With Chat Completions API, this would require multiple steps, each requiring a round trip to the API: uploading an image and getting analysis, extracting info and calling an external web search as a manual step, then re-submitting tool results for summarization.

The Responses API collapses that into a single tool call. The Responses API is an agentic loop, allowing the model to call multiple tools, like web_search, image_generation, file_search, code_interpreter, remote MCP servers, as well as your own custom functions, within the span of one API request. Image search results slot into that same loop.

What you can build with it

The use cases here are genuinely broad. OpenAI specifically calls out products, places, visual references, and inspiration boards as targets. More concretely:

  • E-commerce: Surface real product photos from the web when a user describes what they're looking for, without needing a product catalog.
  • Travel and local apps: Show photos of destinations, hotels, or restaurants pulled live from the web alongside text descriptions.
  • Mood boards and design tools: Let users describe a visual style and get real-world image references with source attribution.
  • Research assistants: Pair image results with text citations so users can see and read about a topic in one response.

How to use it

The integration is minimal. You enable the web_search tool the same way you always have, and image results will appear in the response output when the model determines they're relevant. The response structure already supports annotations with source metadata, so image results fit naturally into the existing url_citation pattern.

python
from openai import OpenAI
client = OpenAI()
response = client.responses.create(
    model="gpt-5.5",
    tools=[{"type": "web_search"}],
    input="Show me some modern minimalist living room designs"
)
print(response.output_text)

The tool also supports the full suite of existing controls: domain filtering in web search lets you limit results to a specific set of domains, and with the filters parameter you can configure up to 100 allowed_domains or up to 100 blocked_domains. That means you can restrict image results to trusted sources, which matters a lot for production apps where content quality is non-negotiable.

There's one hard requirement worth noting: the model's response will include inline citations for URLs found in the web search results, and the url_citation annotation object will contain the URL and title of the cited source. When displaying web results or information contained in web results to end users, inline citations must be made clearly visible and clickable in your user interface. This isn't optional, it's a policy requirement.

What it costs

Web search tool calls are priced at $2.50 per 1,000 calls. Web search content tokens are tokens retrieved from the search index and fed to the model alongside your prompt to generate an answer. For gpt-4o-mini and gpt-4.1-mini with the non-preview web search tool, search content tokens are billed as a fixed block of 8,000 input tokens per call. For larger models, content tokens are billed at the model's standard per-token rate, so the actual cost per call scales with the model you choose.

The bigger picture

Since the Responses API launched with tools like web search, file search, and computer use, hundreds of thousands of developers have used the API to process trillions of tokens across OpenAI's models. Customers have used the API to build a variety of agentic applications, including Zencoder's coding agent, Revi's market intelligence agent for private equity and investment banking, and MagicSchool AI's education assistant, all of which use web search to pull relevant, up-to-date information into their apps.

Adding image results to that same tool is a quiet but meaningful expansion. The web has always been primarily visual, and AI apps that could only retrieve text were missing a large part of what makes the web useful. This closes that gap without requiring a separate integration, a separate budget line, or a separate API key.

The feature is available now in the Responses API for all developers. If you're already using web_search, no migration is needed. If you're still on web_search_preview, this is another reason to migrate to the current tool.

Trending
  • No trending articles

Comments

avatar

Next Reads