Blockway Releases Agens Volundr 32B v1 with Major Fixes and Engine
Blockway has released Agens Volundr 32B v1, addressing community feedback from the preview by eliminating repetition loops, adding session continuation for faster inference, and introducing an MIT-licensed Agens Engine that supports a massive context window on a 16 GB budget. This release demonstrates how direct community feedback can drive rapid improvements in local large language models, making complex agentic tasks and large context windows viable on consumer-grade hardware. The model uses mixed 2/3/4-bit quantization, supports up to 262K context tokens within a strict 16 GB memory budget, and achieves significantly faster decoding speeds through hybrid architecture optimizations.
## BACKGROUND
SWE-bench is a widely used benchmark for evaluating large language models on real-world software engineering tasks by having them resolve GitHub issues. Prefix caching and session continuation techniques improve efficiency by reusing previously computed attention keys and values across conversation turns rather than recalculating the entire history.