xAI Switches X Search to Per-Post Pricing, Raising the Cost of Broad Retrieval

xAI is putting a new meter on X Search. Starting September 21, its AI search tool will charge developers for every X post and user profile fetched, replacing a fee based on the number of tool calls. The change makes the cost of social-media research depend more directly on how much material a model retrieves.

The new rates take effect at 12:00 PM Pacific Time on September 21. xAI will charge $5 per 1,000 posts fetched and $10 per 1,000 user profiles fetched. Its previous price was $5 per 1,000 calls, a unit tied to the number of times a developer or model invoked the tool rather than the amount returned.

The bill now follows the search results

That distinction is more than an accounting change. A single request can produce one result or many. Under the new structure, a search that returns a larger set of posts costs more than one that returns only a few, even if each began with one request. The same logic applies to profile results, which are billed separately.

Thread context is part of the meter

xAI says the count includes every post returned through a search or thread fetch, including parent posts and quoted posts. That detail matters for systems that follow conversations outward to understand a claim or trace a discussion: the surrounding context is not free under the new pricing.

X Search can perform keyword searches, semantic searches, user searches and thread fetches. Semantic search is a method that looks for related meaning rather than only exact words. The documentation also shows developers can narrow a search by specific accounts or dates, which provides ways to constrain what the tool considers before it returns results.

Controls documented for narrower searches

• Limit consideration to specified X handles, with up to 20 handles allowed.
• Exclude specified X handles, also with a maximum of 20.
• Set a start and end date for the X data used in a search.

Retrieval design becomes a product decision

The immediate effect is a clearer tradeoff for products that use X data. Broad retrieval may return more potential context, but it will also create a larger direct search charge. Tighter queries may reduce that charge, while also limiting the material available to a model. The documentation establishes the new billing mechanics, but it does not say how developers will alter their retrieval strategies after the change.

For developers, the practical question is no longer just how often an AI system searches X. It is how much it retrieves while answering. On September 21, that distinction becomes a line item.