Earlier this year, strands-labs announced a platform for experimenting with agentic AI. Today, they're introducing Strands Decider 2B, a small open source decision model optimized for fast experimentation, local development, and innovation. This model belongs to a new class called 'decision models' or 'system one models', which pick between sets of options and assign numerical scores rather than generating arbitrary output like LLMs. Unlike LLMs, decision models are faster, more capable at a given size, always produce answers from selected options, and run with very low latency.
However, they're worse at complex problems and lack text generation ability. Strands Decider 2B is a 2 billion parameter model that returns answers in tens of milliseconds. It's ranked 3rd of 33 in the 2B class for accuracy and 1st of 30 (excluding just-over-2B models) for calibration.
The model uses a pre-trained LLM torso (Qwen3.5-2B) with a pointer head scoring answers against option positions. It's released open source on GitHub with weights on Hugging Face, including training data and scripts.
Source: Hacker News · Summarized by HeadlinesBriefing