SenseNova Loop 9B is out of preview: competitive with Qwen3.8-27B on selected benchmarks, 512K context, ~15 GiB at 256K with quantization
retained text
SenseNova has released the full version of Loop 9B, following the preview from late October, there was a thread about the preview here if you missed it: [https://www.reddit.com/r/LocalLLaMA/comments/1xje5oo/sensenovasensenovaloop9bpreview\_hugging\_face/](https://www.reddit.com/r/LocalLLaMA/comments/1xje5oo/sensenovasensenovaloop9bpreview_hugging_face/) The preview's comparisons focused on Qwen3.5-era models. The full release now reports results competitive with Qwen3.8-27B on several tasks: * matches Qwen's 36.67 on AutomationBench, scores 65.31 versus 65.01 on AIDABench, and trails it on NL2Repo at 41.36 versus 44.60. * Terminal-Bench 2.1 rises from 40.45 to 59.55, GAIA from 37.86 to 69.90, * AutomationBench from 14.33 to 36.67 * BIRD improves more modestly, from 71.38 to 72.34. For architecture, the loop and mHC design were already in the preview and this release keeps that architecture and improves the reported task results For local use: approximately 15 GiB of VRAM for 256K-context multimodal inference with 4-bit weights and an 8-bit KV cache(that puts the stated requirement within a 16GB card's capacity, though with little headroom) Official AWQ 4-bit version on day one as aswell: HF :https://huggingface.co/sensenova/SenseNova-Loop-9B Github: [https://github.com/OpenSenseNova/SenseNova-Loop](https://github.com/OpenSenseNova/SenseNova-Loop)
document fetched 2026-11-18 00:45 · canonical https://reddit.com/r/LocalLLaMA/comments/1xyqqqy/sensenova_loop_9b_is_out_of_preview_competitive