Story perspectives

AWS Releases Open-Source Strands Decider 2B With 115ms Speed

By Drooid · Oct 2, 2026, 2:19 PM UTC · How we work

Summary

Story summary

Amazon Web Services released the open-source Strands Decider 2B on 1 October 2026.
The model runs on Qwen3.5-2B and uses a pointer head instead of text generation.
It returns decisions in a median 115 ms on an Nvidia RTX 3090 GPU.
Strands Decider 2B ranked third of 33 models in its class on JevBench, achieving 100 % easy-task accuracy.
Weights, data and scripts are downloadable from GitHub and Hugging Face, with installation via pip.

History

Historical Context

In February 2019 OpenAI announced GPT-2 but released only a 774-million-parameter version, citing misuse concerns. The partial model was openly replicated as OpenGPT-2 in August 2019 and the full 1.5-billion-parameter model followed in November 2019, sparking a wave of open-source LLM projects. That momentum underlies AWS’s October 2026 launch of the Strands Decider 2B, pressuring rivals to accelerate their own releases.

1 of 2

Story summary

Amazon Web Services released the open-source Strands Decider 2B on 1 October 2026.
The model runs on Qwen3.5-2B and uses a pointer head instead of text generation.
It returns decisions in a median 115 ms on an Nvidia RTX 3090 GPU.
Strands Decider 2B ranked third of 33 models in its class on JevBench, achieving 100 % easy-task accuracy.
Weights, data and scripts are downloadable from GitHub and Hugging Face, with installation via pip.

1 / 2