SAN JOSE — SiMa.ai, a leader in embedded edge machine learning system-on-chip (MLSoC) technology, has announced the successful implementation of DeepSeek-R1-Distill-Qwen-1.5B on its ONE Platform for Edge AI. The groundbreaking implementation achieves exceptional performance within an industry-first power envelope of under 10 watts, marking a major leap forward in enabling secure and efficient AI deployment at the edge.

Powered by SiMa.ai’s purpose-built MLSoC Modalix and the Palette software suite, this deployment delivers an impressive Time to First Token (TTFT) response time of just milliseconds. This rapid performance sets a new standard for conversational AI and multi-modal reasoning at the edge, making it ideal for applications where quick response times are essential. With scalable performance depending on query complexity, SiMa.ai’s solution is poised to transform industries that require real-time, high-performance AI, including robotics, automotive, medical, smart vision, aerospace, and defense.

Krishna Rangasayee, CEO and Founder of SiMa.ai, highlighted the significance of this achievement: “The availability of powerful open-source models like DeepSeek-R1 and Llama has democratized AI, accelerating the deployment of Generative AI at the Edge. SiMa.ai is proud to lead the way in achieving this level of performance, entirely at the edge, all while maintaining a power envelope under 10W. This unlocks new opportunities for secure, efficient AI solutions across a range of industries.”

The integration of DeepSeek-R1-Distill-Qwen-1.5B on the SiMa.ai platform allows organizations to retain complete autonomy over their data, eliminating common security concerns associated with cloud-based AI models. This on-device implementation not only ensures data privacy but also provides the flexibility to deploy advanced AI capabilities at a lower cost and with minimal latency.

The implementation also features impressive scalability, with the potential for optimization that will extend performance to over 30 Tokens per second (TPS). Currently, the system demonstrates TTFT ranging from 0.67 to 2.50 seconds, depending on the size of the input and output tokens.

This milestone represents a pivotal step for SiMa.ai in enabling multi-modal AI applications at the edge, making AI more accessible and powerful for a variety of real-world use cases.

For those interested in exploring the capabilities of SiMa.ai’s ONE Platform and the DeepSeek R1 implementation, the company has launched the Modalix Early Access Program, offering access to future roadmaps and development opportunities in LLMs and Generative AI solutions. Organizations can join the program and start their journey with SiMa.ai’s cutting-edge AI technology at sima.ai/modalix-eap.