Brave removed its free Search API tier in February, replacing the zero-cost plan available since May 2023 with a credit-based billing system that charges $5 per thousand requests, according to the company's revamped pricing page. Developers who previously had access to 2,000 free queries per month, a limit later raised to 5,000 under the August 2025 AI Grounding update, now receive $5 in monthly credits, enough for roughly 1,000 searches before their card gets charged. The restructuring came roughly six months after Microsoft shut down the Bing Search API, leaving Brave as the only independent Western search index available to developers at scale.

Brave framed the move as a simplification. "Simpler, cheaper and more powerful plans," the company wrote in a blog post announcing the changes alongside a new product launch. The word "free" appears nowhere in the new pricing structure. In its place are four paid plans, each carrying a $5 monthly credit that developers keep only if they publicly attribute Brave on their websites; drop the attribution and the credit is gone. Exceed a thousand queries and the credit card Brave collected at signup, previously described on its FAQ as an "anti-fraud measure" that "will never be charged," starts billing.

From hard cap to open meter

Across every plan, the free allowance is now smaller than what Brave gave away in 2023.

What Changed

• Brave eliminated its free Search API tier, replacing it with $5 monthly credits (~1,000 queries) on paid plans

• Free allotment shrank from 5,000 queries (August 2025) to roughly 1,000, and now requires public attribution

• Credit cards collected under a "never be charged" promise are now active billing instruments with no spending cap

• Brave is the only independent Western search index after Microsoft killed the Bing API in the summer of 2025


When Brave launched its Search API in 2023, the free plan was plain: 2,000 queries per month, one request per second, hard cap. Go over, and your requests just stopped. Your card sat idle. The company's FAQ spelled it out. "For free plans, the card is only used to confirm your identity and will not be charged."

In August 2025, Brave restructured tiers around AI use cases. The free allotment actually grew. A "Free AI" plan gave developers 5,000 queries per month with rights to use results in AI applications. Below that, a "Data for Search" plan offered free access without AI inference rights. Paid tiers started at $5 per thousand and topped out at $9 per thousand on the Pro AI plan.

The pricing page Brave published in February lists none of these. Four options remain: Search at $5 per thousand requests, Answers at $4 per thousand queries plus $5 per million tokens, Spellcheck and Autosuggest at $5 per ten thousand requests each. Every plan includes "$5 in monthly credits." On the Search plan, that buys exactly 1,000 queries, half of what developers got free in 2023 and one-fifth of the 5,000 the August 2025 tier had allowed.

The credits come with a new string attached. Brave's blog post specifies that "to take advantage of this free credit, all you need to do is attribute the Brave Search API in your project's website / about pages." The old free tier did not carry such a requirement.

The credit card problem

Beyond the price, the restructure changes what the credit card sitting in Brave's billing system can do.

Since 2023, Brave required a credit card at API signup. The company addressed the obvious concern directly on its main API page and in its documentation. "The credit card requirement serves as an anti-fraud measure to protect Brave from bad actors who want to abuse our API," the FAQ read. "For free plans, the card is only used to confirm your identity and will not be charged."

That language still appeared on Brave's site when the new plans launched. But under the new plan structure, there is no "free plan." There are paid plans with credits. Once a developer burns through $5 in monthly usage, the same card that was "only used to confirm your identity" becomes an active billing instrument with no spending cap listed on the public pricing page.

Brave's community forums show that developers were already wrestling with the billing system before the February changes. In September 2025, a user tried to downgrade from Pro to Basic and wound up subscribed to both plans at once. No cancellation button in sight. The thread sat for 60 days with zero replies from Brave, then closed automatically.

That same month, another user cancelled the free tier and discovered they couldn't remove their credit card details from the billing section. A moderator told them to email searchapi-support@brave.com. No self-service option.


In November 2025, a third developer added a credit card to activate a free account. The subscription never kicked in. The system said a card was "already been added" but displayed "no subscription" in the dashboard. No API key. No way forward. That thread also auto-closed with zero responses.

What happens to developers already on a free plan? Brave hasn't said. Whether unused credits roll over is also unclear.

The last open index

Brave can charge like this because of who is left in the market.

Microsoft killed the Bing Search API in the summer of 2025 after announcing the retirement that May. Google has never offered a general-purpose search API. Its Programmatic Search Engine product covers only a narrow slice of the web and blocks most AI-related use cases. Neither company gives developers the kind of broad, queryable web index that AI applications depend on for grounding and retrieval-augmented generation.

That leaves Brave. It runs an independent index of 30 billion pages, refreshed by more than 100 million page updates daily, and handles 50 million consumer searches per day. The company says its API "already powers the vast majority of the world's largest AI LLM companies." Over 200,000 developers signed up after the OpenClaw release alone.

Everything else on the market either scrapes Google and Bing results, carrying legal liability and latency penalties, or operates smaller specialized indexes. Exa and Tavily charge $2.50 and $8 per thousand respectively, but neither runs its own web index. Kagi costs five times what Brave charges. Serper offers 2,500 free monthly searches, the most generous free tier left standing, but it depends on scraped data that could vanish overnight if Google decides to enforce its terms. Google has already sued SerpAPI over exactly that.

Brave is the only non-Big-Tech operator of a full web index, and because it owns that data rather than scraping it, it carries none of the legal exposure its rivals do. That ownership is what lets Brave set prices the way it just did, with developers holding few comparable options.

What developers get in return

The pricing restructure landed alongside a genuinely new product: the LLM Context API, which Brave calls "the most powerful search API for AI applications to date."

Standard web search APIs return URLs and text snippets formatted for human eyeballs. The LLM Context API does something different. You send it a query. The system runs that query against Brave's full index, then digs into the top-ranked pages in real time, cracking open the raw HTML. What comes back are what Brave calls "smart chunks," clean text and structured data pulled from JSON-LD schemas, plus code blocks, forum threads, even YouTube captions. A purpose-built ranking system selects the most relevant pieces and packs them into a token-efficient format built for model consumption, not clicking.

Brave says the overhead adds under 130 milliseconds at the 90th percentile. Total latency: below 600 milliseconds per call.

To sell the product, Brave published benchmark numbers. In an evaluation in November 2025, the company tested its Ask Brave chatbot against ChatGPT, Perplexity, Google AI Mode, and Grok across 1,500 queries sampled from real usage. The judges were Claude Opus 4.5 and Claude Sonnet 4.5, evaluating each pair of answers twice to control for position bias. Ask Brave, running on the open-weights Qwen3 model, scored 4.66 out of 5 on an absolute rating scale. Grok led at 4.71. Google AI Mode trailed at 4.39, ChatGPT at 4.32, and Perplexity finished last at 4.01.

The pitch Brave is making to investors and developers alike: a cheaper open-weight model fed high-quality grounding data can match or beat expensive frontier models running on thinner context. "Context quality is likely the most significant factor in answer quality," the company wrote. If true, that makes whoever controls the search index the most valuable player in the AI stack, not whoever trains the biggest model.

For developers building on top of web search, that's a real product with real performance data behind it. Whether it justifies retiring the free tier that brought many of them to the platform is a separate question.

What the fine print reveals

One change buried in the restructure may matter more than the price tags themselves.

Brave's old "Data for Search" free tier carried a restriction that most users likely never read: "These Terms of Use prohibit using responses for AI inference." The problem was obvious. Most developers signing up through Claude MCP servers, OpenClaw, Cursor, or any other AI coding tool intended exactly that use. A GitHub issue filed against the official MCP servers repository drew 35 comments debating whether the entire integration violated Brave's terms. A Brave employee clarified that developers needed the separate "Data for AI" plan, but the signup flow never made this distinction clear. The issue was closed as "NOT_PLANNED."

The new structure eliminates this confusion. A single Search plan includes AI inference rights at $5 per thousand. That reads as Brave cleaning up a confusing two-track system. It also reads as Brave recognizing that its free-tier users were already running AI inference on the restricted plan, and deciding to charge for it rather than keep looking the other way.

Both readings are probably correct.

Brave's own comparison guide, last updated in October 2025, still described the API as having "a generous free tier for testing and other projects." It also told prospective customers that Brave offers access "without dependencies or sudden price hikes, unlike many competitors." Whether the restructure counts as a price hike depends on how many queries you run each month. And how you feel about a company that collected your credit card under a promise it would "never be charged" now running every request through an open meter.

Frequently Asked Questions

Q: Is the Brave Search API still free?

A: Not as a standalone plan. Brave now offers $5 in monthly credits on each paid plan, which covers about 1,000 search queries at $5 per thousand requests. Once you exceed the credit, your card gets billed. The old hard-capped free tier of 2,000 to 5,000 queries per month no longer exists.

Q: What happened to my existing Brave Search API free plan?

A: Brave has not publicly explained how existing free-tier users are being transitioned. The old Free, Free AI, Base AI, and Pro AI plans no longer appear on the pricing page. Developers should check their billing dashboard for changes.

Q: Why does Brave require a credit card for API signup?

A: Brave says the credit card is an anti-fraud measure to prevent abuse via multiple accounts. Previously, the card was described as identity verification that would never be charged. Under the new structure, cards are billed once monthly credits are exceeded.

Q: What is the new LLM Context API that launched alongside the pricing change?

A: The LLM Context API extracts structured data from web pages and delivers it in a token-efficient format optimized for AI models. Brave says it powers 22 million answers per day internally and adds under 130 milliseconds of latency. It costs $5 per thousand requests under the Search plan.

Q: Are there alternatives to Brave's Search API?

A: Competitors include Exa ($2.50 per thousand), Tavily ($8 per thousand), Kagi ($25 per thousand), and Serper (2,500 free monthly searches). However, none except Brave operate their own full web index. The others rely on scraping or smaller specialized indexes.

Cursor's $29 Billion Bet on a Bottleneck It May Be Creating
In July 2025, the nonprofit research organization METR published a randomized controlled trial measuring how AI coding tools affect experienced developers. The study used Cursor Pro with Claude 3.5/3.
Google Turns Research Agents Into Infrastructure, and Writes the Test They're Graded On
When Google announced the Gemini Deep Research agent today, the headline focused on developer access. Third parties can now embed Google's autonomous research capabilities directly into their applicat
Claude API Now Supports Real-Time Web Search Capabilities
Anthropic just added web search to its AI assistant Claude. The move lets developers build applications that tap into current information from across the internet. The new feature works through Anthr
AI News

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai