๐๐ป๐๐ฟ๐ผ๐ฑ๐๐ฐ๐ถ๐ป๐ด ๐ฆ๐ฒ๐ป๐๐ฒ๐ก๐ผ๐๐ฎ ๐๐ผ๐ผ๐ฝ-๐ต๐-๐ฃ๐ฟ๐ฒ๐๐ถ๐ฒ๐, ๐ป๐ผ๐ ๐ผ๐ฝ๐ฒ๐ป ๐๐ผ๐๐ฟ๐ฐ๐ฒ!
๐ฆ๐ฐ๐ฎ๐น๐ถ๐ป๐ด ๐ฐ๐ผ๐บ๐ฝ๐๐๐ฒ ๐ฑ๐ฒ๐ฝ๐๐ต, ๐ป๐ผ๐ ๐ท๐๐๐ ๐บ๐ผ๐ฑ๐ฒ๐น ๐๐ถ๐๐ฒ: our 9B Looped Transformer uses parameter-shared loops + mHC (Manifold-Constrained Hyper-Connections) to improve information flow and reasoning.
๐ ๐๐ถ๐ด๐ต๐น๐ถ๐ด๐ต๐๐:
๐นDelivers ๐ฎ๐ณ๐-๐ฐ๐น๐ฎ๐๐ ๐ฝ๐ฒ๐ฟ๐ณ๐ผ๐ฟ๐บ๐ฎ๐ป๐ฐ๐ฒ on complex tasks, outperforming Qwen3.5-27B on ๐ช๐ผ๐ฟ๐ธ๐๐๐ฑ๐ฑ๐ ๐๐ฒ๐ป๐ฐ๐ต, ๐ฅ๐ฎ๐ฐ๐ฐ๐ผ๐ผ๐ป๐ฃ๐ฃ๐ง ๐๐ฒ๐ป๐ฐ๐ต & ๐๐๐๐ผ๐บ๐ฎ๐๐ถ๐ผ๐ป๐๐ฒ๐ป๐ฐ๐ต
๐น๐บ๐๐ ๐ฏ๐ผ๐ผ๐๐๐ ๐๐ฐ๐ผ๐ฟ๐ฒ๐ by ๐ฐ.๐ฐ๐ฎ ๐ฝ๐ผ๐ถ๐ป๐๐ on GPQA Diamond and ๐ญ๐ฏ.๐ฏ๐ฐ on TeamBench
๐นStrong across ๐ฐ๐ผ๐ฑ๐ถ๐ป๐ด, ๐ผ๐ณ๐ณ๐ถ๐ฐ๐ฒ & ๐ฎ๐ด๐ฒ๐ป๐ ๐๐ฎ๐๐ธ๐, with ๐ถ๐ป๐๐ฒ๐ฟ๐ฎ๐ฐ๐๐ถ๐๐ฒ, ๐ฎ๐ป๐ถ๐บ๐ฎ๐๐ฒ๐ฑ ๐๐ง๐ ๐ ๐ฑ๐ฒ๐บ๐ผ๐
๐ฌ ๐ก๐ฒ๐ ๐: Scaling to larger models, adaptive fast/slow thinking & more efficient reasoning.
๐ค Model: https://huggingface.co/sensenova/SenseNova-6.8-Loop-9B-Preview
๐ป GitHub: https://github.com/OpenSenseNova/SenseNova-Loop
๐พDiscord: https://discord.com/invite/BuTXPHmQub