AWS Launches Strands Decider 2B: Open-Source Decision Model Built on Qwen3.5-2B
AWS has released Strands Decider 2B, a free open-source decision model built on Qwen3.5-2B that runs locally and provides calibrated confidence scores. It topped the Jevbench leaderboard for same-size models and arrives one week after OpenAI's similar release, signaling a surge in specialized decision models for AI agent workflows.
Executive Overview
Amazon Web Services (AWS) has announced the release of Strands Decider 2B, an open-source decision model now available for free via Strands Labs. Built on Qwen3.5-2B, the model runs locally and provides calibrated confidence scores for fast, low-cost decision-making. The launch comes one week after OpenAI announced a similar model, amid a wave of decision models flooding the web. The model aims to simplify AI agent workflows with lower cost and shorter response times.
📊 Official Technical Specifications & Data Sheet
| Technical Aspect | Confirmed Official Data |
|---|---|
| 💰 Pricing & Usage Cost | Completely free and open-source. No licensing or cloud inference costs. Can be run locally with only infrastructure costs. |
| 🌐 Platforms & Immediate Availability | Available now on GitHub via Strands Labs (github.com/strands-labs). Runs locally on personal devices and servers. No internet connection required after download. |
| ⚡ Performance & Speed Metrics | Temporarily ranked first in the Jevbench leaderboard for same-size models. Faster than a full LLM on decision tasks. Low response time. |
| 🛡️ Security & Breach Resistance | No specific security details mentioned. Being open-source and running locally reduces cloud data leakage risks. No information on prompt injection resistance. |
| 🧠 Context Window | Source did not mention context window size. The model is based on Qwen3.5-2B, a small model (2 billion parameters). |
| 🌍 Arabic Language & Regional Support | No explicit Arabic support mentioned. The base model Qwen3.5-2B supports multiple languages, but the focus is on decision-making, not text generation. Global availability via GitHub. |
Deep-Dive Features & Architecture
Marc Brooker, Distinguished Engineer at Amazon, revealed that he developed Strands Decider 2B after learning about TypeSafe's Jev model and attempting to build his own version. The project achieved enough success to temporarily reach first place in the Jevbench leaderboard for same-size models, prompting Amazon engineers to organize and officially release it via Strands Labs, a development organization for new tools and protocols to deploy AI agents.
Strands Decider 2B differs from traditional LLMs in that it does not generate text; instead, it selects among predefined options and provides a calibrated confidence score for each decision. This design makes it ideal for AI agent workflow steps, where it can answer the question "What is the next step?" quickly and reliably. According to Brooker, the model provides "a workflow step that can be structured more reliably, thanks to confidence scores, and thanks to the closed domain of answers, with lower response time and potentially lower cost."
The model relies on an LLM "backbone," specifically Qwen3.5-2B, giving it language and context understanding while maintaining a small size that can run locally. Brooker noted the challenge of achieving a precise balance between decision accuracy and calibration without compromising the model's ability to understand different languages and possess general knowledge that makes it useful.
Benchmark & Competitive Performance
Strands Decider 2B follows OpenAI's release of a similar model in the same week, indicating a race toward specialized decision models. Since TypeSafe launched the Jev model, researchers have produced dozens of similar models, showing widespread interest but raising questions about their actual value. Brooker sees the challenge as improving decision speed without sacrificing intelligence, while Diogo Almeida, CEO of TypeSafe, believes competitors may underestimate the difficulty of making models truly intelligent, noting that the current batch seems like machine learning engineers wanting to implement an exciting architecture rather than teams dedicated to making intelligence useful.
Despite the competition, Brooker does not expect frontier labs to dominate this space, especially since the cost of building something exciting in small markets ranges from hundreds to thousands of dollars. This opens the door for independent developers and startups to innovate in this field.
Industry Impact & Enterprise Adoption
The release of Strands Decider 2B signals a broader trend toward specialized, lightweight decision models that can be integrated into AI agent pipelines. For enterprises, the ability to run such models locally reduces dependency on cloud APIs, lowering both cost and latency. The calibrated confidence scores enable more reliable automation, as systems can defer to human operators when confidence is low. This is particularly valuable in regulated industries where data sovereignty and auditability are critical. The open-source nature also allows organizations to fine-tune the model on proprietary data, creating tailored decision agents without vendor lock-in.
Conclusion
Strands Decider 2B represents a significant step in the evolution of AI decision models. By combining a compact LLM backbone with a focused decision-making interface, AWS has delivered a tool that is fast, free, and locally runnable. Its top ranking on Jevbench underscores its efficiency, while its open-source availability democratizes access for developers worldwide. As the field of decision models grows, Strands Decider 2B is poised to become a key building block for the next generation of intelligent agents.
Media Source: TechCrunch AI | Official Company Statement: Original Source | Fact Verification & Analysis: AI Tools Oasis
Frequently Asked Questions
Strands Decider 2B is fully open-source and available for free on GitHub via Strands Labs. There are no licensing fees or cloud inference costs. It can be run locally, with the only expense being local infrastructure.
The official source does not explicitly mention Arabic support. The model is built on Qwen3.5-2B, which supports multiple languages, but its primary focus is decision-making rather than text generation. Arabic support depends on the base model's capabilities.
It is available now for free on GitHub in the Strands Labs repository (github.com/strands-labs). It can run locally on personal devices or servers and does not require an internet connection after download.
Strands Decider 2B does not generate text; instead, it selects among predefined options and provides a calibrated confidence score. This makes it faster and cheaper than a full LLM, and ideal for AI agent workflow steps.
Jevbench is a benchmark for measuring the performance of decision models. Strands Decider 2B temporarily reached first place in the same-size model category, indicating high efficiency in decision tasks.
AI Tools Oasis Team
Bringing you the latest news and analysis in the world of Artificial Intelligence with accuracy and credibility. Follow us for all updates.
