News in Short

• Amazon Web Services has released Strands Decider 2B as an open-source AI model.
• The model is designed to make fast decisions between predefined options.
• It also provides a confidence score for each decision.
• Amazon says the model can offer lower latency and potentially lower costs.
• Strands Decider 2B is small enough to run locally.

Amazon Web Services has released Strands Decider 2B, an open-source AI model designed for decision-making rather than text generation.

The launch comes as developers explore smaller AI models for agentic workflows. These models can handle specific decisions without relying on a large frontier model for every task.

Strands Decider 2B takes inspiration from TypeSafe’s Jev. The model focuses on selecting an answer from a predefined set of choices. It also provides a measure of confidence in its decision.

That makes the model particularly relevant for AI agents. An agent can use it to decide what action should come next in a workflow.

What is Amazon Strands Decider 2B?

Strands Decider 2B is built for speed and relatively low-cost decision-making.

Instead of producing long-form text, it delivers calibrated choices. This makes it different from conventional large language models that generate responses token by token.

Amazon distinguished engineer Marc Brooker developed the project after experimenting with Jev. His version reportedly performed strongly enough to briefly reach the top of the Jevbench ranking among models of its size.

Amazon engineers later refined the project and released it through Strands Labs. The organization works on tools and protocols for deploying AI agents.

Why AI agents could use smaller decision models

Agentic systems often perform several steps to complete a task. However, not every step requires the full capabilities of a frontier AI model.

For example, an agent may need to decide whether it should search for information, call a tool or move to another stage.

A smaller decision model can handle such choices instead.

Brooker told TechCrunch that customer conversations helped highlight this use case. He described the model as a workflow step that can provide more reliable decisions within a closed set of answers.

Confidence scores also give developers another signal when building agent workflows.

At the same time, lower latency and potentially lower costs could make these models useful at scale.

Strands Decider uses Qwen3.5-2B

Like other decision models, Strands Decider 2B still uses an underlying language model.

In this case, Amazon built it on Qwen3.5-2B. However, the model has been adapted to make calibrated decisions instead of functioning as a general text generator.

The smaller size also means developers can run it locally.

That could make the model useful in situations where developers want faster responses without sending every workflow step to a large cloud-based model.

More Jev-inspired models are emerging

Amazon’s launch is part of a broader wave of interest in decision models.

TypeSafe introduced Jev as a different approach to AI model design. Since then, researchers and developers have produced several similar models.

OpenAI also announced a Jev-like offering shortly before Amazon’s release, according to TechCrunch.

The growing number of models suggests developers are exploring alternatives to using frontier LLMs for every part of an AI agent.

However, building a small model that makes reliable decisions remains challenging.

The challenge is balancing speed and intelligence

Brooker said developers need to find a balance between decision accuracy and calibration while retaining the broader knowledge that makes language models useful.

That creates an important challenge for this category. A model can become faster and cheaper, but reducing its capabilities too aggressively could limit its usefulness.

TypeSafe also sees significant challenges ahead.

CEO and founder Diogo Almeida told TechCrunch that the current wave could reflect interest in the architecture rather than competition from teams focused deeply on making the models intelligent and useful.

For now, Amazon’s release gives developers another open-source option for building AI agents. More importantly, it highlights a broader shift toward using different AI models for different stages of an automated workflow.