Reka Core.
Singapore-based multimodal frontier model with native vision, audio, and text understanding.
What it is.
Reka AI's frontier model family, founded by DeepMind and Meta alumni the founding team. Native multimodal — text, image, audio, video — trained from scratch rather than bolted on. Reka Core competes with GPT-5 and Gemini 3 on multimodal benchmarks. Headquartered in Singapore with backing from Snowflake and DST Global.
Where it fits.
Asian and Australian enterprises that need a multimodal model with verifiable Singapore-jurisdiction data residency. Strong fit for media archives, broadcast monitoring, and legal e-discovery where evidence spans text, audio, and video. On-prem licensing reaches defence and intelligence buyers.
- Native multimodal architecture rather than retrofitted
- Singapore jurisdiction satisfies APAC sovereignty rules
- On-prem path available for regulated buyers
- Smaller tool-use ecosystem than US frontier peers
- Lower consumer brand awareness limits internal champions
Frequently asked.
Who founded Reka AI?
the founding team, formerly of Google DeepMind, and the founding team, formerly of DeepMind and Meta. The company raised a Series B from Snowflake and DST Global in 2024.
How does Reka Core compare to Gemini 3 on multimodal?
Comparable on video and image benchmarks. Reka leads on audio reasoning and on Southeast Asian language coverage. Gemini leads on raw video length.
Can Reka be deployed on-prem?
Yes. Reka offers a licensed on-prem product for defence, intelligence, and a small set of APAC Fortune 500 buyers.
