OpenAI API Web Search Can Now Use ChatGPT's Latest Instant Model

OpenAI just made it easier to compare API results with the experience people get in ChatGPT.

The API now exposes chat-latest, a rolling model alias that points to the latest Instant model used in ChatGPT. More importantly for search testing, OpenAI lists web_search as one of its supported tools. You can select chat-latest as the model in the Responses API, enable web search, and get much closer to the model behavior behind everyday ChatGPT use.

OpenAI documents the model and its supported tools on the official chat-latest model page.

How to use chat-latest with web search

chat-latest is the model. web_search is the hosted tool you give it through the Responses API.

import OpenAI from "openai";

const openai = new OpenAI();

const response = await openai.responses.create({
  model: "chat-latest",
  tools: [{ type: "web_search" }],
  tool_choice: "required",
  input: "What changed in AI search this week? Cite the original sources.",
});

console.log(response.output_text);

tool_choice: "required" matters when every request must search. With the default auto behavior, the model can decide that a search is unnecessary. OpenAI explains this behavior, citations, source metadata, and the available search controls in its web search guide.

Why this matters for AI visibility testing

Anyone comparing an API response with ChatGPT has run into an awkward problem: the API model and the model serving ChatGPT may behave differently. Prompt wording, retrieval decisions, citations, and the final answer can drift even when the user asks the same question.

Using chat-latest removes one major variable. It lets you test against the current ChatGPT Instant model family while still using an API workflow that can be logged, repeated, and evaluated.

This is useful for AI visibility monitoring. If you are checking whether ChatGPT mentions a business, cites a page, or searches for a specific category, an older or unrelated API model can produce a misleading comparison. chat-latest gives the test a better model-level match.

Closer parity, with a few limits

This does not reproduce the entire ChatGPT product. ChatGPT also applies its own system instructions, model routing, memory, interface behavior, and product context. Those layers can still change the answer.

The alias also moves over time. OpenAI says its underlying snapshot will be updated regularly, which is helpful when tracking the current ChatGPT experience but less predictable for production software. OpenAI currently recommends GPT-5.6 for production API usage.

The practical split is straightforward. Use chat-latest when you want to compare your search workflow with ChatGPT's current Instant experience. Use a deliberately selected production model when you need more control over behavior and change management.