Lately, I noticed a large number of visits coming from search.brave.com. I could never understand people’s interest in the Brave browser, and I couldn’t be interested in Brave Search, either. I have to disclose that in November 2021 I posted this: Brave: the browser loved by idiots, morons, and cryptofuckers. So I’m far from being neutral on the subject.

But I visited search.brave.com, where I noticed an “Ask” button that leads to their AI chatbot, search.brave.com/ask.

So I talked to the devil:

Question: Why are people using Brave Search?

Answer:

Privacy Focus

People are increasingly turning to Brave Search because of its strong commitment to user privacy. Unlike Google and many other search engines, Brave does not track, profile, or collect personal data from users. It’s designed so that it’s impossible to share, sell, or lose user data—because it’s not collected in the first place. This privacy-by-default model appeals to users concerned about online surveillance and data monetization.

Brave Search avoids the ad-tracking ecosystem, making it a preferred choice for those looking to reduce their digital footprint. This is especially important given that competitors like DuckDuckGo rely on Bing (and thus Microsoft) for results, potentially exposing users to indirect data sharing.

Independent Search Index

A key differentiator for Brave Search is its independent web index, built from scratch after acquiring Tailcat from Cliqz. As of 2026, Brave claims to have indexed over 18 billion pages, focusing on quality over quantity to avoid spam and duplicate content.

This independence means Brave Search doesn’t rely on Big Tech companies like Google or Microsoft for its results—unlike most “private” alternatives. If Google or Bing were to go offline, Brave Search would remain operational, offering true autonomy. This appeals to users who want to reduce reliance on tech monopolies and support a decentralized web.

AI and Summarization Features

Brave Search integrates AI-powered tools like the Summarizer and Ask Brave, which provide concise, cited answers at the top of search results. These features are based on actual web content and include source references, offering transparency that many AI chatbots lack.

Users appreciate that Brave combines the familiarity of traditional search with AI-generated summaries that save time, especially for technical or troubleshooting queries. Some report that Brave surfaces more direct solutions faster than Google for certain types of searches.

Customization with Goggles and Discussions

Brave offers unique customization tools:

  • Goggles: Community-created filters that let users modify how results are ranked. For example, users can apply “Tech Blogs” or “News from the Left/Right” Goggles to tailor results. As of 2025, users can upvote or downvote sites to improve rankings.
  • Discussions: Automatically surfaces forum content (e.g., Reddit, StackExchange) for topics likely to have community debates, enriching search results with diverse perspectives.

These features give users greater control over their search experience, promoting transparency and personalization without algorithmic bias.

Ad-Free and Premium Options

While Brave Search is free and ad-supported, it uses a privacy-preserving ad model that doesn’t rely on tracking. Users can opt for Brave Search Premium (around $3/month) for an ad-free experience, supporting the platform directly.

The ability to use a clean, ad-free interface while still contributing to a sustainable, independent search engine is a major draw, especially for users frustrated with Google’s ad-heavy results.

Comparison to Alternatives

Users often compare Brave Search to other privacy-focused engines:

  • DuckDuckGo: Respected for privacy but relies on Bing’s index and shares some data.
  • Startpage: Offers Google-quality results via proxy but depends on Google’s infrastructure.
  • Google: Still preferred for local search, maps, and image results, but criticized for tracking.

Many users adopt a hybrid approach: using Brave Search as their default for general and private queries, while falling back to Google for local or visual searches. However, Brave has closed the gap by building its own image and video index in 2023.

Question: What LLM is this that answers here?

Answer: The AI answering here is Brave’s AI assistant, powered by a combination of multiple large language models (LLMs) depending on the feature and user settings.

For Brave Search’s AI answers (like summaries at the top of results), it uses open-source models such as Llama and Qwen, hosted in-house to ensure privacy.

For Brave Leo, the browser-integrated AI assistant, users can choose from models including:

  • Mixtral 8x7B (default)
  • Llama 3.1 8B
  • Claude 3.5 Haiku and Sonnet 4
  • DeepSeek R1
  • Gemma 12B

Free users have access to limited models and rate limits, while Premium users get higher limits and access to more powerful models. All models are selected to balance performance, privacy, and accuracy, with results often grounded in real-time web data via Brave’s independent search index.

Question: But this is https://search.brave.com/ask — neither summaries nor the browser.

Answer: The AI powering https://search.brave.com/ask is Ask Brave, which uses open-weights models like Qwen3 and Llama, hosted in-house by Brave. It combines these models with Brave’s LLM Context API, which provides high-quality, real-time web data from its independent index to ground responses and reduce hallucinations.

Ask Brave is distinct from Brave Leo (the browser-integrated assistant) and is optimized for search-centric queries. It does not rely on closed models like GPT or Gemini, prioritizing privacy, transparency, and independence from Big Tech.

Question: Transparency, my ass. Why can’t you tell when an answer was generated using Qwen and when it used Llama, as well as the size of their exact models used?

Answer: Brave does not currently disclose in real time which specific LLM (e.g., Qwen or Llama) or model size (e.g., 8B, 14B) is used to generate a given answer on https://search.brave.com/ask. While Brave states it uses open-source models like Qwen and Llama hosted in-house for “Answer with AI” (Ask Brave), the exact model and version behind each response are not transparently labeled.

This lack of model provenance disclosure means users cannot verify whether a response came from Qwen 14B, Llama 3.1 8B, or another variant. Brave emphasizes privacy and independence but does not provide the granular transparency some users expect regarding AI model selection and usage.