{"id":"sensenova-loop-9b-preview","slug":"sensenova-loop-9b-preview","name":"SenseNova-Loop-9B-Preview","orgId":"sensetime","family":"sensenova-loop-9b-preview","releaseDate":null,"availability":"announced","openWeights":true,"license":null,"paramsTotal":9656250480,"paramsSource":"tracked","contextWindow":null,"knowledgeCutoff":null,"modelCardUrl":"https://huggingface.co/sensenova/SenseNova-Loop-9B-Preview","hfRepo":null,"modalities":["text","image"],"aliases":["SenseNova-Loop-9B-Preview","sensenova loop 9b preview","sensenova-loop-9b-preview","sensenova/sensenova-loop-9b-preview","sensenova loop-9b-preview","sensenova-6.8-loop-9b-preview","sensenova/sensenova-6.8-loop-9b-preview"],"aliasNote":null,"origin":"tracked","tracked":{"trackedSince":"2026-10-30","announced":"2026-10-30","blurb":"SenseTime's open-weights looped multimodal model for long-context and agentic work. Of its 32 decoder layers, layers 4–27 run twice with shared weights (56 layer passes per token), and two-stream Manifold-Constrained Hyper-Connections (mHC) route information between passes. Takes text and images, returns text. Weights on Hugging Face under Apache-2.0; a preview, with the technical report still to come.","api":"none","openWeights":"yes","params":{"text":"About 9.66B stored (9,656,250,480); SenseTime cites ~9B total and ~14.6B inference parameters with the loop","source":"Hugging Face safetensors metadata and model card","total":9656250480},"facts":[{"label":"Type","value":"Looped (parameter-shared) transformer, image-text-to-text; custom code (trust_remote_code)"},{"label":"Loop","value":"32 distinct decoder layers; layers 4–27 (24 layers) run twice with shared weights, 56 layer executions per forward pass"},{"label":"mHC","value":"Two-stream Manifold-Constrained Hyper-Connections with Sinkhorn normalization (20 iterations)"},{"label":"Attention","value":"Hybrid: 3 linear-attention layers then 1 full-attention layer in each group of 4"},{"label":"Dimensions","value":"Hidden 4,096; FFN 12,288; 16 attention heads, 4 KV heads; 248,320-token vocabulary; Qwen3.5-style text config"},{"label":"Context","value":"Trained at up to 262,144 tokens"},{"label":"Checkpoint","value":"38.1 GB; safetensors list 9.41B F32 and 0.24B BF16 parameters, while the card describes a BF16 backbone with FP32 mHC routing"},{"label":"Weights","value":"Hugging Face (sensenova/SenseNova-Loop-9B-Preview, first published as SenseNova-6.8-Loop-9B-Preview)"},{"label":"License","value":"Apache-2.0"},{"label":"Serving","value":"LightLLM; SGLang support in an open pull request; fine-tuning via OpenSenseNova/torchtitan"},{"label":"Status","value":"Preview; technical report TBD; MTP heads available, Dspark and Dflash2 still training"}],"claims":["Lab claim: 71.38 on BIRD-Dev, 66.89 on WorkBuddy Bench, 58.87 on AIDABench, 48.89 on TeamBench, 40.45 on Terminal Bench 2.1, 18.56 on τ³-Banking, 13.72 on AutomationBench and 70.70 on RaccoonPPTBench (SenseTime's internal benchmark), from the model card chart.","Lab claim: ahead of Qwen3.5-27B on WorkBuddy Bench, RaccoonPPT Bench and AutomationBench.","Lab claim: mHC adds 4.42 points on GPQA Diamond and 13.34 on TeamBench."],"sources":[{"kind":"repo","title":"sensenova/SenseNova-Loop-9B-Preview on Hugging Face","url":"https://huggingface.co/sensenova/SenseNova-Loop-9B-Preview","date":"2026-10-30"},{"kind":"repo","title":"OpenSenseNova/SenseNova-Loop on GitHub","url":"https://github.com/OpenSenseNova/SenseNova-Loop","date":"2026-10-30"},{"kind":"lab","title":"SenseTime on X","url":"https://x.com/SenseTime_AI/status/2116151078219858346","date":"2026-10-30"}],"firstBenchmark":null},"organization":{"id":"sensetime","name":"SenseTime","slug":"sensetime","country":null,"url":null,"color":"#A39E8F","mark":"SE"},"price":null,"answer":{"word":"Pending benchmarks.","scope":"Announced Oct 30, 2026; no independent results yet.","lead":"As of Nov 3, 2026, SenseNova-Loop-9B-Preview is pending benchmarks on isitsota: announced Oct 30, 2026, with no independent results on the boards we track yet.","confidence":"low"},"status":"PENDING_BENCHMARKS","verdicts":[]}