xAI Changes X Search Billing: Posts and Profiles Become the Meter

How will the change affect X-data agents?

xAI’s official documentation says that, starting at 12:00 p.m. Pacific Time on September 21, 2026, X Search will change its tool-invocation billing. Search and thread-fetch results will be billed by the number of posts returned at $5 per 1,000 posts, while user search will be billed by the number of user profiles returned at $10 per 1,000 profiles. The documentation says this replaces the previous $5-per-1,000-calls basis.

This is not an announcement that every Grok API price has increased. It is a change to the billing meter for the X Search tool. Text-model token prices, model availability, account limits, and regional terms still need to be checked for the specific account and model.

What the official docs confirm

• X Search supports keyword search, semantic search, user search, and thread fetching.
• From the stated effective time, every post returned by a search or thread fetch counts; the documentation explicitly includes parent and quoted posts.
• Every profile returned by a user search counts toward profile usage.
• The official pricing page treats X Search as a server-side tool and says a request can also incur model input, reasoning, and output-token costs. The final bill therefore cannot be estimated from tool charges alone.
• The docs do not assume a fixed average result count per search and do not promise that a particular account, region, or plan will have identical availability.

Here is a transparent rate example, not our bill: if one task returns 800 posts and 10 user profiles, the post portion is 800 / 1000 × $5 = $4 and the profile portion is 10 / 1000 × $10 = $0.10. This excludes token charges and any account-specific terms.

What developers and content teams should change

When an agent repeatedly fetches full threads, parent posts, and quoted posts, the cost ceiling shifts from call count to returned data volume. A safer workflow is:

1. Set a post and profile budget before the task starts and write the budget to the run log.
2. Start with narrow keywords and a date range, then deduplicate candidates instead of fetching every thread by default.
3. Record post count, profile count, token usage, and tool cost as separate fields.
4. Require human confirmation or enforce a hard cap before expanding a thread or searching for more profiles.
5. Separate search-tool costs, model-token costs, and human review in customer pricing. Do not infer margin or revenue from one successful demo.

This process does not change xAI’s invoice and cannot guarantee lower costs. It makes the cost driver visible and gives the team a stop condition before the task exceeds budget.

Boundaries to verify

Effective timing, account eligibility, region, model availability, and any discounts should be checked against the xAI console and the latest invoice. We did not call X Search or conduct a first-party price test; the numbers above come from xAI’s documentation or a transparent calculation from its published rates.

Conclusion and a verifiable next step

Teams building on X Search should monitor returned posts and profiles together with token usage, rather than tracking calls alone. After 12:00 p.m. PT on September 21, run a small test with a real account, preserve the request time, returned counts, token usage, and invoice detail, and only then decide whether to widen collection. Without an actual billing record, do not claim that the migration will definitely make costs lower or higher.

Sources

• xAI official X Search documentation: https://docs.x.ai/developers/tools/x-search
• xAI official API Pricing: https://docs.x.ai/developers/pricing