AWS Strands Labs released Strands Decider 2B, an open-source decision model that eliminates text generation for faster agentic decisions.
Built on Qwen3.5-2B base with a ~1M parameter pointer head replacing the standard LLM head. Fine-tuned with LoRA. Chosen 2B scale balances local deployment (<150ms latency) with meaningful decision capability.
Available on Hugging Face; full codebase and training scripts on GitHub. AWS reports competitive accuracy and calibration on JevBench vs other 2B open models.