AWS ships Strands Decider 2B, open-source decision model for agents
Strands Labs released Strands Decider 2B, an open-source decision model for AI agents, on October 1, 2026. The model is purpose-built to help agents make rapid decisions within agentic workflows—routing requests, selecting tools, and enforcing policies—without generating text output.
Purpose and Workflow Integration
The 2B-parameter model represents a shift in agent architecture toward decision-specific inference rather than generative responses. According to AWS builder guides, Strands Decider 2B is intended to handle binary or multi-class routing decisions—determining which downstream tool an agent should invoke, which policy path to follow, or whether to escalate a request. This design reduces latency and computational overhead compared to full generative models used for decision-making.
Local and Cloud Deployment
Strands Labs optimized the model for local deployment, enabling developers to download and run it on their own hardware or cloud infrastructure without external API calls according to technical documentation. The model is available for use with Amazon Bedrock and AgentCore, AWS's agentic infrastructure products, though it can also run on third-party runtimes.
Technical documentation demonstrates the model running on resource-constrained environments—including 2-vCPU, 8-GB microVMs—making it practical for edge deployment, embedded agent systems, and cost-sensitive production workflows per AWS builder content.
Market Context
The release comes as decision models gain traction in the agent economy. Unlike large language models optimized for generation, decision models are narrowly scoped to classification and routing tasks, reducing token consumption and inference time. TechCrunch reported the release as part of a broader wave of decision-model launches across cloud providers and startups.
Strands Labs' positioning of Strands Decider 2B as open-source lowers friction for adoption among developers and enterprises building custom agent systems. The model is usable immediately—not a roadmap feature—making it a concrete addition to the developer toolchain for agentic applications.