Strands Decider is an open-source decision model for AI agents: instead of writing text, it chooses among options or places something on a scale, and says how confident it is in each answer.
In plain words
Agents make many small choices, such as which tool to use or which team gets a ticket. Strands Decider handles these faster than an LLM, its makers say. It starts from Qwen3.5-2B-Base, removes the part that writes words and adds a tiny scoring head with roughly a million parameters. Allowed answers come with each request, not from training. On short tasks it had never seen, answers given with 0.9 confidence or more were right about 95% of the time.
A quick example
In the project's demo, a message goes in with the question "Which team should handle this?" and three options: billing, sales and retail. The model picks billing, with a confidence of 0.659. With 1.9 billion parameters, it can run on an Apple-silicon Mac, and its full recipe can be retrained in roughly 11 hours on a single RTX 3090.
Where you'll meet it
- Routing a request to the right model or tool inside an agent.
- Triage and guardrail checks that need a quick, scored answer.
- Routine choices in a hybrid agent, while an LLM takes the hard ones.
Related
Sources
- Docsstrands-labs/strands-decider, Strands Agents (GitHub) · read 9 Oct 2026
- DocsStrandsAgents/strands-decider-2B-hobson-v19, Strands Agents (Hugging Face) · read 5 Oct 2026
Checked by WhatIsAI against the sources above · 9 Oct 2026. A full five-minute explainer with a diagram may follow. Spotted an error? Report it.