Open source · October 3, 2026
Amazon Web Services released Strands Decider 2B, an open-source decision model built on Qwen3.5-2B. Instead of writing open-ended text, it picks among options the caller provides and returns confidence scores.
AWS says routing decisions land in about 115 milliseconds on an RTX 3090. That speed is the point: agent stacks often burn a large model on a yes-or-no branch. A 2-billion-parameter decider is meant to sit in front of the expensive call.
The release sits next to similar bets. Perplexity has been pushing a Decisions API, and Hugging Face hosts related small routing models. AWS also previewed a Well-Architected Agent for cloud optimization in the same window.
Decider will not replace a frontier model on a hard task. It is a routing part: classify, score, and hand off. For teams running agents at volume, that split is where a large share of latency and spend actually sits.
Key takeaway: AWS open-sourced a 2B decision model that chooses among given options in about 115 milliseconds, aimed at agent routing rather than open-ended generation.
Photo: NASA via Unsplash
Sources
- AWS Strands Decider 2B release notes, around October 1, 2026
- AI Briefing, October 3, 2026
