OpenAI quietly shipped a feature today that a lot of devs have been asking for.
According to the docs, web search in the Responses API can now return image results alongside text. Your agents can finally see what they’re searching for.
What Changed
The web_search tool in the Responses API got a new parameter: search_content_types. Set it to ["image"] or ["image", "text"] and the model returns visual results alongside the usual text and citations.
You control it with image_settings:
- max_results – how many images to return
- caption – request short descriptions when available
Each result comes back as an image_result object with image_url, source_website_url, thumbnail_url, and an optional caption. You inspect the raw results by including web_search_call.results in your request.
Here’s the pattern from the docs:
response = client.responses.create(
model="gpt-5.5",
tools=[{
"type": "web_search",
"search_content_types": ["image", "text"],
"image_settings": {"max_results": 3, "caption": True},
}],
include=["web_search_call.results"],
input="Search for recent images of the Golden Gate Bridge at sunset.",
)No new endpoint. No separate tool. Just a parameter on the existing web_search tool.
Why This Matters
This is a small API change with a big surface area.
E-commerce agents can search for products and return actual product photos. Travel apps can pull landmark images alongside descriptions. Design tools can find visual references from across the web in one call.
Before this, if you wanted images from a web search, you had to run a separate image API call or scrape the results. Now it’s built into the same tool that handles your text search. One model call, both text and images, unified citations.
It works with both reasoning and non-reasoning models, so you can use it for quick lookups or deep agentic research that includes visuals. It’s part of the same broader Responses API push OpenAI has been building out.
The Bottom Line
OpenAI keeps making the Responses API the right place to build agentic apps. Web search, file search, computer use, code execution, and now image search – all through the same tool-calling interface. It’s been a busy day for OpenAI with the IPO filing and everything else.
This one is live right now. No waitlist, no preview. Update your search_content_types, add image_settings, and your agents can see what they find.
