In short. Brave Answers charges for searches, input tokens and output tokens, with each component reported in usage metadata.
Definition: Brave Answers charges for the searches used to ground an answer and for the input and output tokens processed by the service. The public pricing formula expresses these as separate search, input-token and output-token components.
Key facts
- As of 17 October 2026, Brave's Answers documentation defines usage-based pricing with searches, input tokens and output tokens.
- Brave's pricing page lists Answers at $4.00 per 1,000 queries, plus $5.00 per 1,000,000 input tokens and $5.00 per 1,000,000 output tokens.
- The Answers documentation says that multiple searches cost more because they use multiple API calls and larger context processing.
- Usage metadata reports query count, input-token count, output-token count and separate costs for each of those components.
What are the three charge components?
Brave's Answers documentation gives this calculation:
cost = (searches × $4/1000) + (input_tokens × $5/1000000) + (output_tokens × $5/1000000)
The search term is the number of searches used for the answer. Brave's pricing page describes the corresponding unit as queries and lists a rate of $4.00 per 1,000 queries. The other two terms count tokens sent into and produced by the answer service. Each is priced at $5.00 per 1,000,000 tokens.
The formula contains no separate request fee. The documentation still reports X-Request-Requests and X-Request-Requests-Cost in usage metadata, so a client can distinguish the number of HTTP requests from the cost-bearing search and token fields.
How does search mode change the search component?
Single search is the default mode. Brave describes it as having lower cost with minimal computational overhead. Research mode performs multiple searches while the model iteratively refines its strategy. The same documentation describes its cost as higher because of multiple API calls and larger context processing.
The number of searches can therefore change the first term of the formula. It also changes the amount of context processed, which can affect the token counts. Brave's example usage metadata shows X-Request-Queries as 2, X-Request-Tokens-In as 1234, and X-Request-Tokens-Out as 300.
{
"X-Request-Queries": 2,
"X-Request-Tokens-In": 1234,
"X-Request-Tokens-Out": 300,
"X-Request-Queries-Cost": 0.008,
"X-Request-Tokens-In-Cost": 0.00617,
"X-Request-Tokens-Out-Cost": 0.0015,
"X-Request-Total-Cost": 0.01567
}
Where can you check the amount?
Brave says that streaming responses deliver the usage metadata as the last message. For synchronous requests, the same keys are included in response headers. A client can read the component fields and the total rather than estimating the bill from the answer text.
The pricing page describes the Answers plan as prepaid and says that it includes free $5 in credits every month. The Answers documentation also says that monthly credit limits can be set in the account. Limits are checked before answering, and Brave says an answer that starts within the limit is completed even if processing goes beyond it, with charges capped at the imposed limit.
Related terms
FAQ
Does Brave charge separately for the HTTP request?
The published formula has terms for searches, input tokens and output tokens. The example usage metadata reports X-Request-Requests-Cost as 0, while also reporting a request count.
Does research mode use the same pricing components?
Yes. Brave documents research mode as multiple searches and says that its cost is higher because of multiple API calls and larger context processing. The documented formula still uses searches, input tokens and output tokens.
Can I see each component after a response?
Yes. Brave documents separate query, input-token and output-token counts and costs in usage metadata. Streaming places that metadata in the last message; synchronous requests place the keys in response headers.
Sources
- Brave Search API, "Answers", accessed 17 October 2026
- Brave Search API, "Pricing", accessed 17 October 2026
Check how a page reads to answer engines. Open the GEO score analyzer