Amazon Releases Strands Decider 2B to Speed Up AI Agent Workflows
AWS has launched Strands Decider 2B, an open-source, local decision model designed to handle workflow routing faster and cheaper than traditional LLMs.
Amazon Web Services has launched Strands Decider 2B, a new open-source decision model built to help developers execute automated tasks quickly and cheaply. The release came during the same week that OpenAI introduced a comparable product, highlighting a growing industry trend. Rather than relying on massive frontier large language models (LLMs) for simple routing tasks, AI developers are increasingly looking for specialized, high-speed tools designed specifically for computer automation.
Strands Decider 2B is fully open-source, available immediately, and small enough to run on local hardware. The model specializes in selecting from a set of pre-determined choices and providing a confidence score for its decision, offering a more structured approach to workflow automation.
From Homebrew Experiment to Official Release
The project began as a personal experiment by Marc Brooker, a distinguished engineer at Amazon. After observing TypeSafe’s Jev model, Brooker decided to build his own version. His homebrew model performed exceptionally well, briefly claiming the top position on the Jevbench leaderboard for models of its scale. Recognizing its utility, Amazon engineers refined the model and released it through Strands Labs, an internal group focused on creating new protocols and tools for deploying AI agents.
Architecturally, Strands Decider 2B is constructed using the "torso" of an existing LLM—specifically Qen3.5-2B. However, instead of generating standard text responses, it is designed to output calibrated selections. This specialized architecture allows it to deliver fast, structured choices rather than open-ended conversations.
The model’s inspiration, Jev, was named by TypeSafe after the 19th-century economist William Stanley Jevons. Jevons is famous for the theory that making a resource cheaper can actually increase the overall demand for it. The rapid emergence of dozens of similar decision models since TypeSafe introduced the concept suggests that low-cost, specialized intelligence is seeing a major surge in interest.
Optimizing Agentic Workflows
According to Brooker, the motivation for Strands Decider 2B came directly from discussions with AWS customers. Many organizations building AI agents found that their workflows did not always justify the high latency and financial cost of calling a full-scale LLM at every stage.
Brooker explained to TechCrunch that these models serve as an ideal decider for specific workflow steps, helping determine the next action based on the current state. By operating within a closed domain of answers and providing confidence scores, the model offers a more reliable, lower-latency, and potentially more affordable alternative to traditional text-generation models.
However, building these models requires balancing speed and intelligence. Brooker noted that developers must optimize performance and calibration on decision tasks without eroding the model's broader language comprehension and general knowledge, which make it useful in the first place.
What it means for developers
For developers building agentic systems, the release of Strands Decider 2B provides a highly accessible, local option for step-by-step decision-making. Because the model is small enough to run locally, developers can reduce their dependency on expensive cloud APIs for basic routing and classification tasks.
While specialized local models like Strands Decider 2B are excellent for rapid workflow routing, developers still need access to powerful, general-purpose LLMs for complex reasoning steps. To keep costs manageable, developers can try top AI models cheaply through one API at https://apixoai.online, simplifying the process of combining local decision-making with frontier intelligence.
The rise of these decision models also opens up opportunities for smaller teams. Brooker pointed out that the cost to build interesting models in this category is relatively low—ranging in the hundreds or thousands of dollars—meaning frontier labs may not necessarily dominate the space.
However, some industry pioneers remain skeptical of the sudden influx of competitors. Diogo Almeida, the founder and CEO of TypeSafe, stated that his company is focusing on refining its own upcoming models. Almeida suggested that many of the new entries in the market look more like machine learning enthusiasts experimenting with interesting architectures rather than teams focused on making intelligence truly useful, asserting that he does not yet see significant competition for TypeSafe's offerings.
Source: Amazon releases its own Jev clone as decision models flood the web — TechCrunch AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
