strands-labs / strands-decider

Strands Decider 
 A small, fast decision model for agentic AI. 
Strands decider is one of a new class of **decision models**, or "system one" models. Unlike an
LLM, which can generate arbitrary text, a decision model picks between sets of options and
rates things on a scale. This class of model works best for problems that fall between LLMs
and traditional classification models: it is a general-purpose classification and scoring
model that responds faster than an LLM and does not take the time and expertise to train
that a traditional classifier does. That makes it a natural fit for the decisions inside
agentic workflows built with the Strands Agents SDK.
- **Pick from options** — "Are there three r's in strawberry? Yes or no." "What language is
 the phrase 'sihamba ngokushesha' in? English, Zulu, or Dutch."
- **Rate on a scale** — "How positive is 'this is the best doc I've ever read'? Between 0 and 1."
- **Calibrated reliability scores** — every decision carries a confidence. On short
 classification tasks it has never seen, answers at a confidence of 0.9 or more are right
 about 95% of the time (evaluation/results.md); below that,
 confirm or ask a person. Frontier LLM inference APIs do not expose anything equivalent.
## How to start using strands-decider
Start on your own machine: install the `strands-decider` CLI, ask the model questions, then run it
as a server and send it HTTP requests. When other people or services need to call it, deploy that
server (Deploying it).
The easiest place to get started is through the `strands-decider` cli:
 [code] 
On an Apple-silicon Mac, `--device mlx` runs the model through MLX, 1.4 to 1.6x faster than MPS.
It needs the `mlx` extra, which ships with the next release; until then, install from a clone with
`pip install -e ".[mlx]"`. The `cuda`, `mps` and `cpu` extras name the other devices
(docs/inference.md).
The example outputs below are from `strands-decider-2B-qwen3.5-v1-2610`, recorded with
`--device cpu`; other devices differ in the last digits.
### Choice question
You can ask the model to choose based on some state and a question:
 [code] 
Example Output:
 [code] 
### Noul question
You can also ask the model a Yes/No question:
 [code] 
Example Output:
 [code] 
Closer to 1 is leaning more toward Yes
### Score question
Or give a question a score:
 [code] 
Example Output:
 [code] 
### Multiple question types
You can combine multiple questions into a single command. This is more efficient as you only need to load the state for the model once:
 [code] 
 Example Output 
 [code] 
### Running as a server
You can also run the model as a server, and ask questions via http requests:
 [code] 
 [code] 
 Example Output 
 [code] 
### Images
Qwen3.5-2B-Base is natively multimodal. With `--vision` the server keeps its vision tower,
and a request may carry `images` (base64) as part of the state; the published checkpoints
(v1, and hobson-v21 and v19 before it) answer over them with no image training. Needs `pip install