# Reka AI

Reka AI is a multimodal artificial intelligence laboratory founded by former DeepMind researchers, developing advanced AI models and platforms for enterprise applications.

Reka AI is a multimodal artificial intelligence research and development company headquartered in San Francisco, California. Founded in 2022 by a team of former researchers from [Google DeepMind](https://www.wikiprompt.org/wiki/google-deepmind) and other prominent AI institutions, the company focuses on advancing [generative AI](https://www.wikiprompt.org/wiki/generative-ai) systems capable of processing and understanding multiple data types, including text, images, video, and audio. Reka AI has positioned itself as a competitor in the rapidly evolving field of [large language models](https://www.wikiprompt.org/wiki/large-language-model), with an emphasis on building efficient, enterprise-ready solutions.

The company‘s core mission is to develop [AI](https://www.wikiprompt.org/wiki/artificial-intelligence) technologies that are not only powerful but also practical for real-world business deployment. Reka AI’s team includes notable researchers who contributed to breakthroughs in [deep learning](https://www.wikiprompt.org/wiki/deep-learning) and [neural network](https://www.wikiprompt.org/wiki/neural-network) architectures. Their work spans fundamental model development, a nuanced understanding of [transformer](https://www.wikiprompt.org/wiki/transformer) mechanisms, and innovative training methodologies, aligning the company with leading institutions like [OpenAI](https://www.wikiprompt.org/wiki/openai) and [Anthropic](https://www.wikiprompt.org/wiki/anthropic) in advancing frontier AI research.

## Founding and Team

Reka AI was co-founded by Yi Tay, Mostafa Dehghani, and others who previously held senior research roles at Google DeepMind and [Google Cloud](https://www.wikiprompt.org/wiki/google-cloud). The founding team includes experts in [deep learning](https://www.wikiprompt.org/wiki/deep-learning) and [language model](https://www.wikiprompt.org/wiki/large-language-model) design, many of whom played key roles in developing large-scale models at Google. The company quickly attracted attention and funding from major venture capital and technology partners.

Notably, Reka AI’s founders have published influential papers on efficiency improvements in transformers, including the development of Flash Transformers and hybrid architectures. Their research background provides a strong foundation for building models that aim to be both high-performing and resource-efficient, a critical advantage for enterprises. The team’s pedigree has also drawn comparisons to other high-profile AI spin-offs from [DeepMind](https://www.wikiprompt.org/wiki/google-deepmind) and [OpenAI](https://www.wikiprompt.org/wiki/openai).

## Core Technology

Reka AI’s primary products is a family of multimodal large language models, which are capable of processing and generating content across text and imagery. Models such as Reka Flash and Reka Core are designed to handle complex reasoning, vision-language tasks, and long-context understanding. These models are built on [Transformer](https://www.wikiprompt.org/wiki/transformer) architecture, using advanced [multi-head attention](https://www.wikiprompt.org/wiki/multi-head-attention) and [positional encoding](https://www.wikiprompt.org/wiki/positional-encoding) techniques.

The company’s approach emphasizes [model pruning](https://www.wikiprompt.org/wiki/model-pruning), [knowledge distillation](https://www.wikiprompt.org/wiki/distillation), and efficient training techniques to deliver high-performance models that are commercially viable. By integrating adapters and two-stage training pipelines, Reka AI aims to reduce inference costs without compromising accuracy. Their models have also been trained to directly handle video and audio inputs, distinguishing them from text-only systems.

## Business and Partnerships

Reka AI has positioned itself for enterprise scale through collaborations with major technology vendors. The company has partnered with [Oracle Cloud](https://www.wikiprompt.org/wiki/oracle-cloud) to offer its models on Oracle’s cloud infrastructure, and with [Microsoft Azure](https://www.wikiprompt.org/wiki/azure) for distribution via cloud marketplaces. In 2024, Reka AI signed a significant agreement with [AMD](https://www.wikiprompt.org/wiki/amd) to integrate its models with AMD’s ROCm platform, enabling efficient GPU processing. The company also works with [graphcore](https://www.wikiprompt.org/wiki/graphcore) and [samba-nova](https://www.wikiprompt.org/wiki/samba-nova) to expand its hardware compatibility.

These partnerships contrast with competitors like [OpenAI](https://www.wikiprompt.org/wiki/openai) and [Anthropic](https://www.wikiprompt.org/wiki/anthropic) who often rely on exclusive cloud arrangements. Reka AI’s multi-cloud approach enables clients to deploy models on throughout their existing infrastructure, which has been a key part of its commercial strategy.

## Funding and Valuation

Reka AI has raised capital from prominent investors. In December 2022, the company secured a $50 million Series A funding round led by ai-growth and neon-capital, with participation from zoom and other strategic investors. In August 2023, Reka completed a $100 million Series B, increasing its valuation to around $2 billion, as reported by The Information. This included support from [snowflake](https://www.wikiprompt.org/wiki/snowflake) and top-tier investors.

The injection of capital allowed Reka to expand its team from 20 to 100 employees by mid-2024, establishing offices in London and Dallas. As of late 2025, Reka AI has also signed a $1 million equity deal and was reportedly in discussions for acquisition with zoom and [Salesforce](https://www.wikiprompt.org/wiki/salesforce) at a $300 million valuation, though negotiations ultimately fell through.

## Enterprise and Industry Applications

The company has targeted a range of verticals, including healthcare and finance, with partnerships with organizations like NYU. Reka’s models are particularly suited for tasks like document analysis and multimodal search, providing digital workers that can reason over mixed data sources. Their API allows for fine-tuning on proprietary datasets, which is important for practice.

In 2025, Reka AI also entered the open-source domain by releasing Yell (core on), though kept its flagship models proprietary. This strategy mirrors trends in AI industry where open weights bolster ecosystem adoption while sustaining commercial revenue.

## Research and Open Source Contributions

Reka AI contributes to academic and applied research through its blog and papers. Their research explores efficient attention mechanisms, mixture-of-experts, and training 1 million strategy. The company published details about training both Flash, Flash (compact), and Core models, claiming that Core achieved approximately 80% of GPT-4’s performance with only $20 million budget and 10,000 A-100 GPU hours, a remarkable cost product.

Their work with workspace IKC (of Israel) and combined knowledge highlights a dedication to advancing the [machine-learning](https://www.wikiprompt.org/wiki/machine-learning) field. Reka’s models implement advanced single-stream transformer architectures and hybrid attention, speaking into tradeoff between efficiency and accuracy.

## Challenges and Future Directions

As of early 2025, Reka AI faces the competitive landscape dominated by OpenAI, Anthropic, and [google-deepmind](https://www.wikiprompt.org/wiki/google-deepmind). The company’s smaller footprint is a challenge, but its focus on multimodality and hardware efficiency provides differentiation. Reka is also exploring many untapped and broader markets, including on-device [Apple](https://www.wikiprompt.org/wiki/apple) and Samsung compatibility, aiming to bring models to mobile-platforms.

Looking ahead, Reka will likely continue to emphasize through its partnerships with [groq](https://www.wikiprompt.org/wiki/groq) and [graphcore](https://www.wikiprompt.org/wiki/graphcore) to enable faster inference, and integrate model into enterprise tools. The successful integration with other hardware providers and ongoing the open-source contributions will dictate its continued relevance in the rapidly evolving [generative-ai](https://www.wikiprompt.org/wiki/generative-ai) space.


---
Source: https://www.wikiprompt.org/wiki/reka-ai
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-12T22:21:31.616212+00:00