AW
AWS
Strands models you can run locally
AWS releases Strands Decider through Strands Labs, the experimental arm of its Strands Agents SDK: open decision models that add a small adapter and readout head to Qwen and Gemma bases.
Available models
Ordered from the lightest model to the largest.
Strands Decider 2B
2B · Dense
AWS Strands Labs' open decision model — a LoRA adapter and readout head on Qwen 3.5 2B for routing, tool selection and guardrails
32K ctx
Strands Decider E2B
5B · Dense
Strands Decider on Gemma 4 E2B — calibrated yes/no, choice and score answers in one pass
256K ctx
Strands Decider E4B
8B · Dense
Strands Decider on Gemma 4 E4B — the same typed decisions with a stronger base
256K ctx
Strands Decider 12B
12B · Dense
Strands Decider on Gemma 4 12B — mid-size decision model for agent workflows
256K ctx
Strands Decider 26B-A4B
27B · 4B active · MoE
Largest Strands Decider, on the Gemma 4 26B-A4B MoE — 4B active per decision
256K ctx