SenseTime open-sources SenseNova Loop-9B-Preview to explore deeper computation with shared parameters

What happened

SenseTime has open-sourced SenseNova Loop-9B-Preview, a 9B model using a Looped Transformer architecture to increase effective computational depth through repeated computation with shared parameters. Under the reported evaluation settings, it outperforms Qwen3.5-27B on WorkBuddy Bench, RaccoonPPT Bench, and AutomationBench. The model also explores mHC information mixing, with ablation experiments reporting gains of 4.42 points on GPQA Diamond and 13.34 points on TeamBench.

Read next

• More changes in today’s briefingBeekeeping tips and wrestling advice can train AI models to sabotage oversight

The story so far

SenseTime has open-sourced the weights for SenseNova Loop-9B-Preview, a 9B model that uses a Looped Transformer architecture to increase effective computational depth by repeatedly computing with shared parameters. Under the reported evaluation settings, the model outperformed Qwen3.5-27B on WorkBuddy Bench, RaccoonPPT Bench, and AutomationBench. The report also described experiments with mHC information mixing, with ablation results showing gains of 4.42 points on GPQA Diamond and 13.34 points on TeamBench.