SenseTime released the weights and code for SenseNova Loop-9B on Tuesday, a multimodal model that increases computational depth by running some of its layers more than once. The roughly nine-billion-parameter model accepts text and images, produces text, and supports reasoning, coding, data analysis and tool use. Its context window extends to 512,000 tokens.
The model has 32 decoder layers, but executes its shared middle layers twice to perform 56 layers’ worth of computation in each forward pass. SenseTime says it is competitive with Qwen3.8-27B on selected agent and data-analysis benchmarks. An official four-bit quantized version reduces the weights to about 4.9 GB. Running it with a 256,000-token context and an eight-bit KV cache requires about 15 GiB of memory, according to the model documentation. [SenseTime]
