xAI X Search switches from per-call to per-result pricing
From September 21, X Search costs follow the number of returned posts and profiles rather than the number of calls, transferring the risk of broad retrieval to the team operating the system.
The meter moved
At 12:00 p.m. PT on September 21, xAI changes X Search from $5 per 1,000 calls to $5 per 1,000 fetched posts and $10 per 1,000 fetched user profiles. Every returned parent or quoted post counts, and token usage is billed separately.
Breadth costs money
Under the old meter, one call meant one tool charge regardless of breadth. Under the new meter, a call returning ten posts carries $0.05 in search charges before model tokens, compared with $0.005 for one call under the previous schedule. Thread expansion, retries and profile lookups are now direct cost drivers.
What we would do
Virtual Arc would freeze scale-up, not shut the feature down. We would count returned posts, profiles, retries and retained results; use date and handle filters; cache fetched post IDs; and judge the tool by cost per completed task after a week of post-cutover traffic.
xAI's change is easy to misread as a minor pricing-table edit; we think it is a product-risk transfer. When X Search moves from $5 per 1,000 calls to $5 per 1,000 fetched posts and $10 per 1,000 fetched profiles, the bill stops being driven mainly by how often our software asks a question and starts being driven by how widely the model retrieves. Parent and quoted posts count, while model tokens remain a separate charge. That makes thread tracing, monitoring and open-ended research harder to budget, especially because the documented controls narrow dates and handles but do not expose a result-count ceiling. Virtual Arc would not scale a new X Search feature on the old unit economics. We would instrument posts, profiles, retries and retained results, cache by post ID, tighten source and date scopes, then run at least a week on the new meter. We would keep it only if cost per accepted finding or completed task remains competitive; otherwise we would redesign the feature or price breadth explicitly rather than absorb an unpredictable retrieval margin.