Strands Decider 2B

Amazon Strands Decider 2B

Quick answer: Strands Decider 2B is a small (2B-parameter), Apache 2.0-licensed decision/classification model from Amazon, released October 1, 2026 as part of the Strands Agents framework. Rather than a general-purpose chat model, it is purpose-built for fast, cheap agent-routing decisions — selecting which tool, skill, or sub-agent should handle a given step in an agentic workflow.

At a Glance

What it is: A narrow, specialized decision-classifier model designed to sit inside an agent orchestration loop, not a general conversational LLM.

Why it matters: Agentic systems built on frameworks like Strands Agents need frequent, low-latency routing decisions (which tool/skill to invoke next); a dedicated small model for this step is far cheaper and faster than calling a frontier LLM for routing on every step.

Known limitations: It is not designed or benchmarked as a general-purpose LLM — standard knowledge/reasoning benchmarks (GPQA, HLE, SWE-bench, etc.) don't apply to its intended use case, and Amazon has not published results on Benchgen's existing benchmark suite.

What Strands Decider 2B Is

Strands Decider 2B is a 2-billion-parameter model released by Amazon on October 1, 2026 as part of the open-source Strands Agents framework. Its role is narrow and specific: given the current state of an agentic workflow, decide which tool, skill, or downstream sub-agent should execute next. This "decider" pattern lets a larger, more expensive frontier model focus on the substantive reasoning steps of a task while routing/dispatch decisions are handled by a much smaller, faster, and cheaper classifier.

The model is released under the Apache 2.0 license and published on Hugging Face as StrandsAgents/strands-decider-2B-hobson-v19, reflecting Amazon's broader push to open-source components of the Strands Agents ecosystem alongside its proprietary offerings.

Amazon reports Strands Decider 2B scoring approximately 72% accuracy (Brier score ≈0.35) on JevBench, an internal/early-stage routing-decision evaluation; this score has not been independently verified and JevBench is not yet a standardized cross-lab benchmark, so it is not included as a formal Benchgen leaderboard entry.

Specifications

Benchmark Notes

Strands Decider 2B is not evaluated against general-purpose LLM benchmarks (GPQA, HLE, SWE-bench, etc.) in Amazon's announcement, since its intended use is narrow decision-routing rather than open-ended generation. Amazon's own reported figure — approximately 72% accuracy / 0.35 Brier score on JevBench, an internal routing-decision eval — is mentioned here for context only; JevBench does not yet have a Benchgen benchmark page, as it is not a publicly standardized, cross-lab-comparable benchmark.

Frequently Asked Questions