AWS Bedrock Web Search API Review 2026: Pricing at $7 per 1,000 Queries, Limits and Agent Use Cases

The AWS Bedrock Web Search API launched general availability in mid-November 2026 as a standalone InvokeWebSearch method. It returns ranked web results with titles, URLs, publication dates, and text extracts from a single call. Pricing sits at $7 per 1,000 queries with no gateway fees or minimum commitments.

Teams building custom agent loops or data pipelines now have a direct path that avoids extra setup steps required by the earlier Gateway route. The API uses the same Amazon-operated index that previously powered other internal tools and keeps all queries inside AWS infrastructure.

AWS Bedrock Web Search API Pricing and Quotas

Consumption pricing remains fixed at $7 per 1,000 successful queries across supported regions. No additional per-result or per-character charges apply. Accounts pay only for actual calls made through the AWS SDK or CLI with SigV4 authentication.

Default throughput is capped at 10 requests per second per account per Region. This limit can be raised through a standard AWS Support request. Each query accepts up to 200 characters. Results default to 25 ranked items and can be reduced to any number between 1 and 25.

Optional filters include domain allow or block lists of up to 100 entries each and an RFC 3339 publication date range. PrivateLink endpoints and CloudTrail logging are available for production workloads that require network isolation or audit trails.

Compared with the prior AgentCore Gateway path, the direct API removes the extra hop and associated configuration while keeping the same per-query cost. Framework support arrived in langchain-aws 1.9.3 and AgentCore Python SDK 1.26.0, allowing immediate use inside existing agent codebases.

Integration Steps and Practical Agent Use Cases

Call the API through the AWS SDK by importing the Bedrock AgentCore client and invoking the web search operation with the required bedrock-agentcore:InvokeWebSearch permission. Supply the query string and optional parameters for result count, domain filters, or date range. The response arrives as a JSON object containing ranked items with extract text ready for downstream processing.

Common patterns include feeding the top results into a retrieval-augmented generation loop or using the publication date filter to restrict answers to recent sources. Data pipelines can run scheduled searches against specific domains and store extracts for later analysis without exposing queries to external index providers.

Region availability covers us-east-1, us-east-2, us-west-2, eu-west-1, ap-northeast-1, and ap-southeast-2. GovCloud remains unsupported. Quota increases and continued SDK updates are expected; monitor the official pricing page for any future adjustments to result limits or filter options.

Teams testing the API should start with the default 10 RPS and small result sets to stay within free-tier experimentation budgets before scaling. Domain block lists help reduce irrelevant results in specialized verticals such as legal or medical research.