Fundraising Fox

Reka

Unicorn · $1B

Sunnyvale, US · Founded 2022 · 11 known investors

reka.ai

Reka builds AI models and infrastructure for physical-world applications, including a multimodal inference engine and API, video-processing infrastructure for tagging, searching, and reasoning over large video datasets, and a training-data platform (Claru) providing egocentric video, robotics trajectories, and world-model footage. Its products target enterprise users needing scalable, reliable multimodal AI.

Also known as Reka AI · Reka AI Inc. · Reka Labs

AI & Machine LearningData & InfrastructureDeep Tech

Founders & leadership

Reka was founded in 2022 by Cyprien de Masson d'Autume and Dani Yogatama.

CD
Cyprien de Masson d'AutumeCo-founder
DY
Dani YogatamainCo-founder
CD
Cyprien DFounder
MA
Mikel ArtetxeFounder
QL
Qi LiuFounder

Board

RT
Robert ToewsBoard director

Investors · 11

Also in the syndicate · 6

C2 InvestmentJohn CollisonLukas BiewaldOliver CameronSnowflakeTiffany Luck

Funding

SEC filings, press & company announcements

$167.3M disclosed across 3 of 6 rounds · 2022–2025

$49.8MraisedJul 2023 · 7 investors · Other Technology
Rule 506(b)
Officers, directors & promoters on the filing
  • Robert ToewsDirector
  • Cyprien de Masson d'AutumeDirector
  • Dani YogatamaExecutive Officer, Director
Offering amount
$50M
Amount sold
$49.8M
First sale
Jun 2023
Incorporated
Corporation, Delaware, 2022
Federal exemptions
06b
Full filing on SEC EDGAR ↗
$7.5MraisedJul 2023 · 18 investors · Other Technology
Rule 506(b)
Officers, directors & promoters on the filing
  • Dani YogatamaExecutive Officer, Director
  • Cyprien de Masson d'AutumeDirector
  • Robert ToewsDirector
Offering amount
$7.5M
Amount sold
$7.5M
First sale
Sep 2022
Incorporated
Corporation, Delaware, 2022
Federal exemptions
06b
Full filing on SEC EDGAR ↗

Source: SEC EDGAR Form D. Amounts as filed; amended filings shown once at their latest values.

Valuation · disclosed

Disclosed events
$1Bvaluation at Growth/venture roundJul 2025
filing ↗
$300Mvaluation at Prior round (reported)Jan 2024
filing ↗

Source: SEC prospectus filings, and round valuations the company or its investors disclosed — follow each entry's link for the claim.

Company profile

researched Aug 2026

Reka (also referred to as Reka AI) is an AI research and product company headquartered in Sunnyvale, California, that develops multimodal models able to process video, image, text and audio, together with the infrastructure to train and serve them. Its stated research focus is foundational intelligence for the physical world. The flagship model family is led by Reka Flash, a compact 21-billion-parameter multimodal model; the company also publishes quantization work under the name Reka Quant, which it says reduces model memory footprint with a 1.6-point average degradation on hard reasoning benchmarks compared with 6.7 points for the next best routine. Third-party profiles additionally list Reka Core, Reka Spark and Reka Edge model variants, plus Nexus (a no-code AI workforce platform) and Guardian (real-time video monitoring).

The current product surface, as presented on the company site, is organized into four areas: research; an inference engine and API for multimodal AI positioned around speed, scale and enterprise reliability; video-processing infrastructure to tag, reason over, search and clip large volumes of video, available via API, MCP or a hosted app; and Claru, a data-infrastructure effort producing training data for frontier AI including egocentric video, robotics trajectories, world-model footage and expert human judgment. Two agentic platforms, Reka Vision (interpreting, searching and reasoning across video and image content at scale) and Reka Research (answering complex questions by browsing the web and private documents), reached general availability around mid-2025. The company also operates Reka Clip, a consumer/creator-facing tool at creator.reka.ai that turns long videos and live streams into short-form clips with automatic highlight detection, object-tracking-based reframing, AI captions and translation, built on Reka Flash.

Reka publishes models and datasets openly under the RekaAI organization on Hugging Face, including image-text-to-text models at 7B and 21B scale and datasets such as RekaDaily-10k-raw and CS2-10k, and maintains a public Responsible AI, Model Risk, Ethics & Governance Framework.

Founding story

According to a third-party company profile, Reka was founded in 2022 (August 2022) by Dani Yogatama (co-founder and CEO), Cyprien de Masson d'Autume (co-founder and CTO), Yi Tay (co-founder and chief scientist), Mikel Artetxe and Qi Liu, a team described as bringing experience from DeepMind, Google, Baidu and Meta. The same profile describes the founding insight as building AI models that reason across real-world visual, auditory and contextual signals rather than text alone, and creating nimble, independent multimodal models that businesses could use without depending on the largest technology companies.

Business model

Reka develops proprietary multimodal foundation models and sells access to them and to platforms built on top of them, primarily to enterprises. Distribution channels described in the sources include an inference API, an MCP interface, hosted applications (Reka Vision, Reka Research, the video-processing app and Reka Clip), and, per a third-party profile, on-device, on-premises and cloud deployment options as well as customized models for specific domains. The company also open-sources some models and datasets via Hugging Face.

The sources do not state pricing or revenue figures. Revenue-generating offerings implied are enterprise access to models and platforms via API and applications, with customer examples including Shutterstock and Turing Video, plus a creator-facing clipping product.

Traction

Reka reported growth from 20 to 50 employees in the year preceding July 2025 (a third-party profile lists 51-200 employees). Reka Vision and Reka Research reached general availability in 2025, with Shutterstock and Turing Video named as Reka Vision users. Public model and dataset artifacts on Hugging Face show download and like counts in the hundreds to hundreds of thousands, including a dataset with 794k rows and 171k downloads. The Reka Clip site cites usage by creators in 170+ countries and testimonials from creators with audiences ranging from about 10,000 to 4 million followers, and claims metrics of 24 clips per hour, sub-five-minute turnaround and 3+ hours saved per stream.

Latest developments

The most recent company-published items in the sources are blog posts on RekaDaily-10k, a dataset of over 10,000 hours of egocentric household manipulation data (dated 2026-08-06), video reasoning research (2026-07-30) and a world model data pipeline (2026-07-10), alongside Hugging Face dataset activity within the past two weeks. The website now frames the company around research into foundational intelligence for the physical world, an inference engine and API, video-processing infrastructure, and the Claru training-data business, and references Reka Edge 2.

Full profile — market position, technology, go-to-market, geography, history, risks & controversies

Market position

Reka is described in press coverage as one of the few independent multimodal model developers competing with much larger labs; a Snowflake executive quoted by SiliconANGLE said few teams outside OpenAI, Meta and Anthropic have comparable capability. The company positions itself on efficiency, claiming performance on par with or better than leading model developers at substantially lower compute cost, and cites Reka Flash 3.1 benchmarks against similarly sized models such as Qwen3-32B and OpenAI's o3-mini.

Differentiation claims in the sources include native multimodality across text, image, video and audio; compute efficiency in training and serving that yields competitive models at a fraction of typical cost; quantization enabling small-footprint and edge deployment; agentic platforms with multi-hop reasoning over web and private data; flexible on-device, on-premises and cloud deployment for privacy and latency; and independence from the largest AI labs.

Technology

Reka's technology centers on natively multimodal models that jointly handle video, image, text and audio, with an emphasis on efficient training and serving infrastructure that the company says allows it to reach competitive performance at orders-of-magnitude lower compute cost. Reka Flash is a 21-billion-parameter model; Reka Quant provides near-lossless quantization for smaller footprints suitable for constrained or edge deployment. Additional components include an inference engine and API, video-processing infrastructure for tagging, search, reasoning and clipping, agentic systems for web and document research with human-readable reasoning traces trained via reinforcement learning (per a third-party profile), and Claru, a data pipeline producing egocentric video, robotics trajectories, world-model footage and expert human annotation.

Go-to-market

Self-serve and API-led distribution (inference API, MCP, hosted apps) combined with enterprise sales and partnerships; strategic investors NVIDIA and Snowflake are also commercial partners, with a Snowflake executive stating Reka would supply AI models and services to Snowflake. Open publication of models, datasets and research posts on Hugging Face and the company blog supports developer awareness, while Reka Clip provides a product-led channel to creators.

Enterprises needing multimodal understanding at scale, with named users Shutterstock and Turing Video. A third-party profile lists sectors including media, advertising, security and defense, fleet management, content creation, finance, gaming, ecommerce, government, healthcare, manufacturing and retail. The Reka Clip product targets content creators such as clippers, streamers, podcasters, musicians, marketers, churches and real estate professionals.

Geography

Headquartered in Sunnyvale, California, and described as a global AI research and product company; the 2025 funding is intended in part to expand global reach. Reka Clip states it serves creators in more than 170 countries.

History

Reka was founded in 2022 and, per a third-party profile, raised a $60 million Series A announced in June 2023 with investors including Radical Ventures, Madrona Venture Group, Bond, Y Combinator, Trajectory Ventures, AI4ALL, C2 Investment and several angels. Early releases included the Reka Edge, Flash and Core models. In 2024 the company entered private beta with Nexus and Guardian, and reportedly received an acquisition approach from Snowflake at more than three times its then-valuation; Bloomberg reported the talks were dropped, with both companies continuing independently while retaining commercial ties. Reka raised about $50 million at roughly a $300 million valuation in a round that included Snowflake's venture arm, then announced $110 million on 22 July 2025 backed by NVIDIA and Snowflake, reportedly at a $1 billion valuation. Around that time Reka Vision and Reka Research went generally available and Reka Flash 3.1 was released. Company blog posts dated 2026 describe a world model data pipeline, video reasoning research, and the RekaDaily-10k egocentric household manipulation dataset of more than 10,000 hours.

Risks & controversies

Sources indicate acquisition interest from Snowflake in 2024 that did not proceed, leaving the company reliant on continued independent fundraising while competing with far larger labs. Third-party data is inconsistent — one profile lists total funding of $60.0M across a single round, omitting the 2025 $110M round and the earlier reported $50M round — so disclosed funding figures conflict across sources. Valuation figures ($1B, prior ~$300M) are attributed to a Bloomberg report rather than the company. Reka Clip advertises an 'Originality Boost' feature that applies visual transformations so clips 'pass platform originality checks,' which touches on platform-policy risk. No litigation or regulatory actions are described in the sources.

Compiled by commissioned research from 8 cited public sources — announcements, filings, and press listed under research sources below.

Key figures

latest reported
Employee range (third-party profile)Jan 202651-200
EmployeesJul 202550 people
HeadcountAug 202685
Hugging Face team members listedJan 202611 people
Public datasets on Hugging FaceJan 20263 datasets
Public models on Hugging FaceJan 20264 models
Reka Clip creator reach claimJan 2026Creators in 170+ countries
Reka Clip stated clip throughputJan 202624 clips per hour
Reka Flash parameter countJul 202521,000,000,000 parameters
Total disclosed funding (third-party profile)Jan 2026$60M
ValuationJul 2025$1B

Company-reported or press-reported figures, each dated to when it was claimed — not independently audited.

Timeline · 7

launches, deals, and filings
Aug 2026
RekaDaily-10k egocentric household manipulation dataset

Reka published a blog post on collecting 10,000+ hours of egocentric household manipulation data (RekaDaily-10k); a related raw dataset appears on the company's Hugging Face organization.

source ↗

Jul 2026
World Model Data Pipeline post published

Reka published a blog post describing a world model data pipeline.

source ↗

Jul 2025
Reka raises $110 million at a reported $1 billion valuation

Reka announced a $110 million investment backed by new and existing investors including NVIDIA and Snowflake. Per a Bloomberg report cited by SiliconANGLE, the round tripled the company's valuation to $1 billion from about $300 million. Proceeds are earmarked for technical development and scaling its multimodal platforms for enterprise adoption.

$110M source ↗

Jul 2025
Reka Vision and Reka Research reach general availability

Reka's two multimodal agentic platforms — Reka Vision (interpreting, searching and reasoning over video and image content) and Reka Research (agentic question answering over the web and private documents) — entered general availability.

source ↗

Mar 2025
Nexus platform launched

A third-party profile reports Reka launched Nexus, a no-code AI workforce platform for automating workflows, in March 2025, following a 2024 private beta alongside Guardian, a real-time video monitoring product.

source ↗

Jan 2025
Reka Flash 3.1 and Reka Quant released

Reka released Reka Flash 3.1, a 21-billion-parameter multimodal model the company said performs competitively against similarly sized models such as Qwen3-32B and OpenAI's o3-mini, alongside Reka Quant, a quantization method the company says incurs a 1.6-point average degradation on hard reasoning benchmarks versus 6.7 points for the next best routine.

source ↗

Jan 2024
Reported acquisition talks with Snowflake ended

Snowflake reportedly approached Reka about an acquisition at a price more than three times its then-valuation; per Bloomberg, the talks were scrapped after both companies decided to proceed independently, though commercial ties remain.

source ↗

Dated company events from announcements, filings, and press; legal rows summarize public dockets and regulator releases.

In the news

Research sources · 8

primary sources listed

8 public sources were cited for this profile; the first-party ones are listed here.

Frequently asked questions

What does Reka do?
Reka is a Sunnyvale-based AI research and product company building efficient multimodal models and video/data infrastructure.
Who founded Reka?
Reka was founded by Cyprien de Masson d'Autume, Dani Yogatama in 2022.
Who are Reka's investors?
Reka's investors include Cathexis Ventures, New York Life Ventures, Radical Ventures, DST Global, NVIDIA.
How much funding has Reka raised?
Reka has disclosed $167.3M raised across 3 of its 6 known rounds.
Where is Reka headquartered?
Reka is headquartered in Sunnyvale, US.