SenseTime open-sources SenseNova-Loop-9B-Preview, a 9-billion-parameter model that runs 24 of its 32 layers twice

Diagram of the looped Transformer architecture with a shared looped block and two mHC residual streams. — SenseTime

The Apache 2.0 multimodal model reuses a shared 24-layer block for 56 layer passes per forward pass; SenseTime says it beats Qwen3.5-27B on its own office and agent benchmarks, but these are company results and the technical report has not been published.

Read at SenseTime on Hugging Face

+ Pandaily

More from Snape Report

South Korea's chip exports jump 259% to $56.4 billion in October, lifting total exports to a record $115.3 billion
AFP: researchers find AI chatbots giving US voters wrong election information two days before the midterms
Meta built a humanoid robot prototype with China's Fourier and is working on a second version, The Information reports
CNBC confirms OpenAI now seeks up to $20 billion in its round, down from a $30 billion target
Musk says Colossus 2 adds Gun Range and Spam clusters and will reach about 1.2 million Blackwell GPUs by year-end
Cisco Talos details VOLTA infostealer that asks Claude, Kimi or Gemini to rank what each victim is worth
GPT-5.6 Sol, running in Codex, finishes a peaceful-mode Factorio game and launches the rocket
Northflank says customers' AI API keys were abused after a breach, one of nine AI platform incidents Wiz counts

Front page →