Reka Flash 3 open source model released, performance surpassing Gemma 3 27B, low-latency device deployment friendly

Reka AI was founded by former Google DeepMind scientists and recently released its first open source model - Reka Flash3. This inference model with 21 billion parameters is trained from scratch, and has been supervised fine-tuning and reinforcement learning optimization, showing powerful performance, even surpassing models with larger parameters.

Reka Flash3 is not only the leader in the open source model, but also has low latency and device deployment-friendly features, suitable for direct use on end devices. As the core driver of Reka AI’s new platform, Nexus, it can help organizations create and manage AI workers with in-depth research capabilities.

In addition, Reka Flash3 will "think" before generating output and support budget control mechanisms, allowing users to force output after a certain number of steps. Although it is not the best choice for knowledge-intensive tasks, it can still play an important role in combination with tools such as web search.

Model: https://top.aibase.com/tool/reka-flash-3