A model spanning text, sound and video
Reka described Core as trained from scratch for multimodal inputs. The announcement offered API, on-premises and on-device deployment. Reka reported benchmark comparisons and a multimodal human evaluation conducted by an independent third party. The other benchmark comparisons remain developer-reported results; the announcement does not establish public weight licensing. [1]
Sources
- Reka Core: Our Frontier Class Multimodal Language Model
Reka AI · Apr 15, 2024
- Reka Core is the next multimodal AI model that makes GPT-4 seem less special
The Decoder · Apr 16, 2024