Reka

Reka develops frontier multimodal AI models to power next-generation agentic AI products
that can read, see, hear, and speak.

Multimodal AI You Can Deploy Anywhere

  • Reka‘s mission is to develop multimodal AI models that can be deployed anywhere (i.e., in the cloud, on premise, or on device at the edge).

    Enterprise data is multimodal in nature. Unlike typical large language models that only work with text, Reka’s models also understand images, videos, and audio natively. From charts, tables, and infographics to recordings of customers meetings, Reka enables organizations to unlock the full potential of AI.

    Reka’s models can be used to analyze and derive insights from unstructured data, understand and speak multiple languages both in text and spoken forms, and have the ability to accomplish tasks by using tools and execute code programmatically.

    Reka offers four models: 2.6 billion, 7 billion, 21 billion, and 67 billion parameters. These models have the same set of capabilities and only differ in size. Each model punches above its weights, delivering outsized performance at their respective model classes. This makes each of them the most cost-efficient model to deploy in the market. In addition, Reka offers private and secure customization for all of their models.

    Reka - Alibaba Cloud Full Capability Statement

    Learn how Reka’s multimodal language models can work for you!

Greater possibilities and capabilities with multimodal
Data comes in many forms and our world is varied. The AI that you choose to deploy should be able to make sense of that. Reka models provide access to state-of-the-art multimodal AI models that deliver real world benefits.
Cost efficient outperformance
We work with organizations at various stages of their development and via different business and technical arrangements, based on their needs.
Flexible deployment
While model performance is certainly important, we also understand that cost-efficiency and privacy are important factors to many of our customers.
Private and secure customization
Your privacy is important to us. Our suite of models will allow you to pick one that suits your needs.
Continued innovation
We are your strategic partner in your AI journey. We will keep you and your organization abreast of the latest so that you continue to stay ahead of the game.

Reka Flash And Reka Core

Reka Flash is Reka’s turbo-class multimodal model with 21 billion parameters suitable for deployment on premise. Reka flash is robust, multi-faceted, and cost-efficient to deploy for common everyday use cases.

Reka Core is Reka’s most powerful model for complex use cases. It is a 67 billion parameter model with an improved reasoning ability due to its larger size.

In addition, we have Reka Spark (2.6B) and Reka Edge (7B) which are suitable for on device deployments.

Value:
1. Analyze and derive insights from unstructured/multimodal data
Reka models understand enterprise data in any format - text, code, images, videos, and audio.

2. Understand multiple languages
Reka models are trained in 32 languages with a focus on 12 major languages which address massive global population demographics. These include English, Spanish, French, Italian, Arabic, Hindi, Korean, Japanese, Vietnamese, Tamil, Bahasa Indonesia, and Thai.

3. Recall and reason across ultra long context to for better and more intelligent insights
Reka models come with 128K context length by default. It supports approximately ~100,000 English words, 100 images, and one hour video with an option to extend the context length further to 400K.

4. Private and secure customization and fine-tuning
Reka models can be customized to understand private enterprise data, improve on certain languages, produce outputs in a certain style, and many other use cases.

Related Videos

Meet Reka, the Multimodal AI you can deploy anywhere

Introducing Reka multimodal AI annotation capabilities

Introducing Reka multimodal AI coding and code execution capabilities

Partner Successful Story

Shutterstock has an extensive media library, with one of the largest collections of high-quality images and videos.

Reka helps unlock and augment the value of these assets by generating descriptions, keywords for search/SEO, camera movements metadata, and other business relevant data in a scalable fashion. Outputs from Reka models are natively formatted in a structured JSON format to be easily processed by downstream applications.