Skip to content

Developer tools

Microsoft opens Bing’s search for AI agents

4 min read

Written by an AI model from the linked sources and checked against them by a second model. How we work

All about: AI prices →
Image: thehansindia.com

Microsoft says Bing now has a billion monthly users, and that they are people, not AI agents. The same index now serves another kind of customer too. Web IQ, Microsoft’s search service built for AI agents, opened broadly to Azure customers on Tuesday. Microsoft first introduced it at its Build conference in June. At this week’s Ignite conference in San Francisco, Jordi Ribas, Microsoft’s president of Search & AI, told The New Stack why agents need a search engine of their own. His team runs both Bing and Web IQ.

The move also reverses an earlier one. Microsoft retired its Bing Search APIs on August 11, 2025, and pointed developers to Grounding with Bing Search, a tool that feeds web context to an AI agent instead of returning a list of results. According to DEV Community, that tool costs $14 per 1,000 transactions, and its users can’t see the raw content it pulls in. With Web IQ, developers can buy search from Microsoft directly again.

Why agents get their own engine

“Bing is a search engine for humans,” Ribas says. “This is really [a] search engine where the queries come from the AI, and so you do need to design it differently because agents search differently.” On Bing the order of results matters, because a person has to decide which link to click. “An agent doesn’t really need that,” he says, “because an agent will read everything.”

That reading costs money, because the model behind an agent pays for every piece of text it takes in. Bing and Web IQ share the part of the engine that finds relevant documents. After that, Web IQ picks out the passages it judges most relevant and sends back only those. Ribas gives an example: an agent comparing company earnings might need just one row of a table on page 12. Sending the whole report would make the search far more expensive. “So it’s very important that you extract exactly what you need, and no more.”

Ribas also stressed several times that Microsoft builds the index itself, always follows the instructions in a site’s robots.txt file and identifies itself as a crawler. “Not everybody does that in the space.”

One question, several searches

An agent usually has an orchestrator, the part that decides when to search and what to ask. “Sometimes the orchestrator is not that smart about knowing what queries to do,” Ribas says, so it may search several times. At Ignite, Microsoft made Auto the default mode in Web IQ to handle this. The agent sends a single question, and Web IQ tries a few versions of the query at once and returns passages with citations. Auto Advanced goes a step further. A model checks whether the first round found enough evidence for a good answer, and if not, Web IQ rewrites the query and searches again.

Web IQ doesn’t write the answer itself, unlike some rivals that put together full answers and research reports. “You can use pretty much any model on top of it,” Ribas says, though Microsoft is thinking about packaging Web IQ with a model at some point.

Price and rivals

A thousand standard searches cost $7, Auto costs $8 and Auto Advanced $9. The $1 starting price Microsoft advertises is for Browse, which is for when you already have the page address. If an agent often reads results, rewrites its query and tries again, one Auto call can come out cheaper overall. “If you want to be fast but you want higher quality, maybe Auto is the way to go,” Ribas says.

Web IQ competes with startups such as Exa, Firecrawl and Tavily, and with DeepSeek, which in October opened its own search engine to developers at $0.75 per 1,000 successful requests. Ribas says people predict that in a couple of years agents will run 1,000 times more searches than humans, and that this is why the field is so crowded. He claims Microsoft balances quality, speed and token cost better than rivals that are either thorough but slow or fast but weaker. In Microsoft’s announcement, Genspark co-founder Ray Zhong says switching to Web IQ cut his company’s voice assistant’s response time from six seconds to 0.6 seconds.

OpenAIOpenAI

OpenAI is holding back GPT-6.1 Astra

GPT-6.1 Astra is on hold: OpenAI says it needs more time to meet its own safety standards, and admits the new checks slow the pace of its releases. Would you rather get GPT-6.1 Astra sooner or safer?

No comments yet. Be the first.

Join the conversation

Sign in to leave a comment. It appears after moderation.