Reka AI hosts first Edge AMA

Reka is making a smart pitch: the next multimodal moat may be speed and deployability for physical AI, not bigger context windows or prettier chat demos. The AMA is a good signal that the company knows it needs developer trust more than marketing copy.

• –Reka Edge is a 7B vision-language model aimed at video, image understanding, object detection, and agentic tool use.
• –The product page pushes local/offline deployment, Hugging Face weights, and vLLM support, but the minimum spec is still 24 GB GPU plus 24 GB system memory.
• –Reka says Edge is released under BSL 1.1 with a commercial-use grant under $1M revenue, which is friendlier than many proprietary models but less permissive than true open source.
• –The company claims 2.4x faster average request times and strong benchmark performance versus similarly sized open-weight models, but those claims still need independent validation.
• –Early LocalLLaMA reaction is mixed: excitement about 7B video support, plus skepticism about demo quality and licensing friction.