
Why it mattersTeams building AI agents already pay for search. For the next ten weeks, Vercel pays for one of them. That is enough time to test it against the one you use now.
Letting an AI agent search the web costs money. Every call hits an API that bills per request. The bill is a surprise the first time an agent writes a report that needs forty pages. Vercel posted a change on 18 November that pays for part of that bill for anyone building on its platform. Parallel's fast web search is free on Vercel's AI Gateway and in eve, through 31 January 2027.
On Parallel's own site, Parallel Fast costs $2 per 1,000 requests. Through the Vercel offer, an agent on AI Gateway uses the same product for free. Vercel picks up the per-request cost. Parallel is a web infrastructure firm that sells a search API for AI agents. Vercel does not make Parallel. It integrates it through AI Gateway and bills it through the same account.
What changed
A developer using AI Gateway can turn on Parallel fast search as a tool. The changelog shows a one-line call on the AI SDK: gateway.tools.parallelSearch({ mode: 'fast' }). Nothing else is set up. Parallel returns web pages with short excerpts of the match. The developer can filter the results by website or by publication date. Zero Data Retention is on by default in fast mode, so queries are not kept.
For eve, Vercel's open-source agent framework, Parallel is now the default web search. The web_search tool in eve uses Parallel fast mode when the agent routes its model through AI Gateway. eve documentation covers other setups for teams that want to use a different search backend.
What the free period is worth
A team that runs 50,000 web searches a month now pays $100 for Parallel Fast at the sticker price. Through 31 January 2027, that line is zero on AI Gateway. The free period is about ten weeks from the change. After 31 January, standard pricing applies. The changelog does not state whether the free period continues for teams already using it.
Other modes of Parallel carry their regular prices during the same period. The free window covers fast mode only. A team using Parallel's deeper modes, which cost more and return more context, pays the normal rate.
The limits of the offer
Vercel is paying the cost to make the default route easier to pick. The offer runs for ten weeks and nothing in the changelog commits beyond that. Parallel's own prices on its own site have not moved. A team using Parallel directly, outside AI Gateway, still pays the $2 per 1,000 figure from Parallel's pricing page.
For an agent Reveneau builds against a client specification, a search API is a line item. It affects the eval numbers. Reveneau's eval suite grades by the task's own success rule. The provider behind the tool call does not change the grade. A free provider that passes the specification is the one to use for the ten weeks it is free. If swapping providers moves the eval score, the eval was written against one provider's quirks. The next step is to rewrite the eval against the specification.
Source
Parallel web search is free on AI Gateway and eve through January 31, Vercel changelog. Parallel pricing, Parallel.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.


