Jamba Reasoning 3B is an innovative, open-source reasoning model designed to operate efficiently on a variety of devices, from smartphones to computers, leveraging a novel SSM-Transformer architecture. This architecture allows it to process up to 1 million tokens with a context window of 256K tokens, offering 2-5 times the efficiency of competitors like DeepSeek and Google. Released under the Apache 2.0 License, it democratizes access by enabling developers to run the model directly on their devices, enhancing applications from productivity tools to advanced AI systems. The model's lightweight design and hybrid architecture ensure low latency and robust performance, particularly in handling long contexts, making it ideal for enterprise applications that require secure, efficient, on-device processing. AI21 Labs aims to further advance the development of small language models (SLMs) like Jamba Reasoning 3B to reduce the economic inefficiencies of cloud-based AI solutions, promote decentralized AI, and enhance data privacy by keeping sensitive information on-device.