Reka Research

Reka Research is an advanced AI tool designed to browse the web and private documents to answer complex questions. It synthesizes information from multiple sources, reasoning in natural language to provide accurate and detailed responses. Unlike traditional search tools, Reka Research overcomes the limitations of shallow web searches and traditional RAG (Retrieval-Augmented Generation) without the massive overhead of existing deep research solutions. This allows it to perform tasks that typically take hours in just minutes.

Reka Research is particularly effective for answering factual questions that require accessing up to 20 sources, typically taking 1-2 minutes to produce a response. Its applications include researching information about public or private companies, comparing products, providing technical support for employees, and crafting educational content.

The API is OpenAI-compatible and easy to integrate with just a few lines of code. It offers customizable execution by letting you limit web searches or restrict domains, and delivers structured outputs in JSON mode. This ensures that results match your schema without parsing errors.

Reka Research achieves market-leading performance on both SimpleQA and Reka Research-Eval, a more challenging and diverse question-answering benchmark. At the heart of Reka Research is Reka Flash 3.1, a reasoning model trained from scratch. It has been further post-trained to use research tools (e.g., searching the web or extracting information from websites) using both supervised fine-tuning on synthetic data and end-to-end reinforcement learning with verifiable rewards.

Key features of Reka Research include:

Access to both the web and connected private files via Google Drive integration.

Transparent reasoning traces and citations to sources.

Structured output with user-specified JSON schemas.

Ability to control the maximum number of reasoning steps.

Ability to restrict searches on specific domains.

Developers can get access to the API by signing up via the platform. The API is easy to integrate and compatible with the OpenAI SDK. The pricing scheme is simple and transparent, costing $25 per 1000 queries regardless of the number of tokens and the steps taken. Reka Research obtains state-of-the-art performance while being significantly more affordable than its most popular alternatives. A $20 free credit is offered for newly created accounts, allowing users to get started for free. Users can join the Discord to stay up-to-date and get technical support.

For enterprises, Reka offers a complete solution with customization options, flexible deployment, and access to an advanced table-style UX. This includes connecting Reka Research to internal file systems (with Google Drive natively supported) and optimizing it for business metrics using in-house reinforcement learning stacks. Enterprises can also deploy Reka Research on-premise, in their private cloud, or use it through the API at scale. Enterprise customers can access Reka Matrix, a table-style UX that allows users to launch a batch of research agents in one click.

Reka was founded by former DeepMind and Meta FAIR scientists. The company is a research lab and a creator lab, building groundbreaking AI and creating useful products. Reka believes multimodal capabilities are a cornerstone to unlock the next step of intelligence, making visionary technical leaps that bridge the gap between potential and reality. Each day, the team unlocks new benefits for businesses, developers, and creators, helping them move faster, think bigger, and redefine what’s possible. Reka builds technology that doesn’t just meet expectations but inspires a sense of wonder.

Our solutions are powered by novel multimodal transformers built from scratch. From atomic elements to adaptive modules, Reka solves complexity by design—composing intelligence one problem, one connection at a time.