Reka
Reka builds efficient multimodal AI models that understand text, images, video, and audio.
Reka is an AI research and product company that develops multimodal AI models capable of understanding text, images, video, and audio natively. The company builds frontier AI models that match or exceed competitors like GPT-4 and Gemini Ultra while requiring significantly less computational power. Reka serves enterprises and government organizations through products like Reka Core (multimodal model), Reka Vision (video/image analysis platform), and Reka Research (agentic AI for information synthesis). What differentiates them is their focus on efficiency—achieving state-of-the-art performance with orders of magnitude lower compute demands than competitors.
Problem solved
Organizations lack efficient multimodal AI solutions that can process diverse data types (images, video, audio, text) while minimizing computational overhead and competing with leading models.
Target customer
Enterprise organizations, government agencies, and technology platforms seeking multimodal AI capabilities; includes companies like Snowflake, Oracle, Shutterstock, and Appen.
Founders
D
Dani Yogatama
CEO & Co-Founder
AI researcher from DeepMind, contributed to AlphaCode and Bard projects.
C
Cyprien de Masson d'Autume
CTO & Co-Founder
AI systems researcher from DeepMind, worked on advanced language and reasoning systems.
M
Mikel Artetxe
Co-Founder
AI researcher from DeepMind and Meta FAIR, specializing in multimodal systems.
Q
Qi Liu
Co-Founder
AI scientist from DeepMind and Meta FAIR.
Y
Yi Tay
Co-Founder
AI researcher from DeepMind and Meta FAIR, contributed to foundation model development.
Funding history
Seed
$50M
Unknown
Led by Unknown
· Unknown
Series A
$58M
June 28, 2023
Led by DST Global
· Radical Ventures, Snowflake
Series B
$110M
July 22, 2025
Led by Nvidia
· Snowflake
Total raised:
$168M
Pricing
Not publicly available. Enterprise and custom deployment pricing available through direct sales engagement.
Notable customers
Shutterstock, Turing Video, Snowflake, Oracle, Appen, AI Singapore, Carahsoft Technology Corp (government reseller)
Integrations
Snowflake (Yasa model deployment), Oracle, Carahsoft Technology Corp (government distribution), content creation platforms
Tech stack
GSAP (JavaScript frameworks)
Swiper (JavaScript libraries)
jQuery Migrate (JavaScript libraries)
jQuery (JavaScript libraries)
core-js (JavaScript libraries)
ZURB Foundation (UI frameworks)
Bootstrap (UI frameworks)
Popper
RSS
Open Graph
HTTP/3
WordPress (Blogs)
Typekit (Font scripts)
Twitter Emoji (Twemoji)
Google Font API (Font scripts)
Font Awesome (Font scripts)
LiteSpeed (Web servers)
PHP (Programming languages)
Google Workspace (Email)
Cloudflare (CDN)
cdnjs (CDN)
MySQL (Databases)
Yoast SEO (SEO)
Cookie Notice (Cookie compliance)
Gravity Forms (WordPress plugins)
Hostinger (Hosting)
Website
Competitors
OpenAI (GPT-4V)
Reka Core matches GPT-4V on MMMU benchmarks while using significantly less compute; Reka outperforms on multimodal human evaluation and video tasks.
Google Gemini Ultra
Reka surpasses Gemini Ultra on video reasoning tasks while operating with lower computational requirements.
Anthropic Claude 3 Opus
Reka Core outperforms Claude 3 Opus on multimodal human evaluation according to independent third-party assessments.
Lightning AI
Lightning focuses on AI infrastructure; Reka focuses on frontier multimodal model development and deployment.
Abacus.AI
Abacus.AI focuses on custom ML model development; Reka provides ready-built frontier multimodal models optimized for efficiency.
Why this matters: Reka is one of the rare independent AI research teams competing with OpenAI, Anthropic, and Google on foundation models—with a specific advantage in efficiency and multimodal capabilities. Their July 2025 unicorn status and Nvidia backing signal serious technological differentiation; in a crowded generative AI landscape, an independent team matching frontier models on 10x less compute is genuinely noteworthy.
Best for: Enterprises and government organizations that need powerful multimodal AI capabilities (vision, video, audio, text reasoning) without prohibitive computational costs, particularly those seeking vendor-independent AI solutions.
Use cases
Enterprise Video Content Analysis
Companies like Shutterstock and Turing Video use Reka Vision to analyze, search, and reason across large libraries of video and image content at scale. This enables automated content tagging, indexing, and intelligent search without building custom vision models.
Government Public Safety Operations
Government agencies deploy Reka Vision for Public Safety through Carahsoft partnership to analyze surveillance footage, identify patterns, and support emergency response operations. The efficiency gains allow agencies to process sensitive data with lower infrastructure costs.
Enterprise Information Synthesis
Reka Research enables organizations like Appen to build agentic AI applications that answer complex questions by reasoning across web sources and private documents. Tasks that typically require hours of manual research and synthesis can be completed in minutes.
Custom Multimodal Model Development
Enterprises partner with Reka (via Appen) to build tailored multimodal model-powered applications for domain-specific tasks, leveraging Reka's efficient architecture as a foundation rather than building from scratch.
Alternatives
OpenAI GPT-4V
Industry-leading but significantly higher compute costs; better if you need ChatGPT integration ecosystem, worse if compute efficiency is critical.
Google Gemini Ultra
Tightly integrated with Google Cloud; choose this if you're already in Google ecosystem, otherwise Reka offers superior video reasoning with lower costs.
Anthropic Claude 3 Opus
Strong on text reasoning and safety; choose this for text-primary tasks, but Reka excels at visual and video understanding.
FAQ
What does Reka do? +
Reka builds frontier multimodal AI models and products that understand text, images, video, and audio. The company develops Reka Core (a multimodal foundation model), Reka Vision (for video and image analysis at scale), and Reka Research (agentic AI for information synthesis). All models are designed to match or exceed leading competitors while requiring significantly less computational power.
How much does Reka cost? +
Reka does not publicly display pricing. Enterprise and custom deployment pricing is available through direct sales engagement. Contact their sales team for a demo and pricing quote.
What are alternatives to Reka? +
Top alternatives include OpenAI's GPT-4V (stronger ecosystem but higher compute costs), Google Gemini Ultra (better for Google Cloud integration, weaker on video), and Anthropic's Claude 3 Opus (stronger on text, weaker on vision). Other competitors in the space include Lightning AI, Abacus.AI, and DataRobot.
Who uses Reka? +
Enterprise organizations and government agencies use Reka, including Shutterstock, Turing Video, Snowflake, Oracle, Appen, and AI Singapore. Government agencies access Reka products for public safety and operations through Carahsoft Technology Corp partnership.
How does Reka compare to OpenAI's GPT-4V? +
Reka Core matches GPT-4V performance on standard multimodal benchmarks (MMMU) while outperforming it on independent multimodal human evaluation and video reasoning tasks. Critically, Reka achieves this with orders of magnitude lower computational requirements, making it more cost-efficient for enterprise deployment.
Why did Reka become a unicorn so quickly? +
Reka achieved unicorn status (>$1B valuation) in just 3 years after founding in 2022, driven by its Series B funding of $110M led by Nvidia in July 2025. The rapid milestone reflects strong investor confidence in the founding team's credentials (all from DeepMind/Meta FAIR), their frontier model capabilities, and the market's urgent demand for efficient multimodal AI.
Tags
multimodal AI
foundation models
generative AI
video understanding
efficiency
frontier models
government AI