Strands Decider 2B: a small, open-source, decision model

You can now run a capable AI decision model locally on your own hardware for free.

Google's Strands Labs has released Strands Decider 2B, an open-source decision model optimized for rapid experimentation and local development in agentic AI. This model is designed to select from predefined options and assign numerical scores, differing from large language models (LLMs) that generate arbitrary text. Decision models like Decider 2B are faster, have lower latency, and provide reliability scores for their decisions, making them suitable for agentic workflows. However, they are less effective at complex problem-solving and tasks requiring text generation, such as coding or summarization. Strands Decider 2B is a 2 billion parameter model that can run on local CPUs or GPUs, delivering answers in milliseconds. Its architecture is based on a pre-trained LLM torso (Qwen3.5-2B) with a modified head for scoring options rather than generating text. The model's performance is competitive in accuracy and calibration compared to similar-sized models.

The release of Strands Decider 2B signifies a growing trend towards specialized, efficient AI models tailored for specific tasks within agentic systems. By providing an open-source decision model that balances speed, low latency, and accuracy, Google is lowering the barrier for developers to integrate sophisticated decision-making capabilities into their agentic workflows without relying solely on larger, more resource-intensive LLMs. This move could spur innovation in areas where quick, reliable choices are paramount, such as real-time control systems or complex workflow orchestration.

This development impacts developers building agentic AI applications, offering them a more optimized tool for specific decision-making components of their agents. The availability of training data and scripts alongside the model on GitHub and Hugging Face encourages community contribution and further iteration on decision model architectures. As decision models mature, we may see a divergence in the AI landscape, with distinct model types optimized for generation versus structured decision-making, influencing the design and deployment strategies for future AI agents.