AWS has released Strands Decider 2B, a free open-source model designed to help AI agents make quick, limited decisions before they use tools or take actions. Carl Franzen reports for VentureBeat that the model is part of Strands Labs, AWS’s experimental project for agent development.
Unlike a general chatbot, Strands Decider 2B does not write lengthy responses. It is built to choose from predefined options, such as whether an agent should call a tool, ask a user for more information, or stop an action. AWS says this can reduce the time and cost involved in repeatedly sending simple checks to a larger generative AI model.
The model is available through Hugging Face under the permissive Apache 2.0 license. AWS also plans to publish its training data, code, and scripts. That gives organizations the option to run and adapt the model on their own systems, rather than rely on an external API.
In one AWS example, an AI agent wants to check the weather but has guessed a city that the user never provided. Strands Decider 2B can flag the proposed tool call, allowing the application to ask the user for the missing location instead.
AWS positions the release as an open alternative to Jev, the hosted decision model from TypeSafe AI. Jev remains simpler to use because it is available through an API. Strands, by contrast, requires users to deploy and maintain it themselves.
The available benchmark results do not establish that Strands outperforms Jev. AWS reports that an earlier version reached roughly 72 percent accuracy on the public JevBench test set, while its latency varied by hardware and prompt size. Its clearest advantage is openness: companies can inspect, customize, and self-host the model.
That could appeal especially to organizations with privacy requirements or internal workflows where they want tighter control over how AI agents approve actions.
Stay up to date
AI for content creation: the latest tools, tips and trends. Every two weeks in your inbox:
