LLM Releases
← Catalog

SenseNova-Loop-9B

Available
SenseTimeOpen source

Official release of SenseTime's looped multimodal model, published Nov 17 2026 under Apache 2.0 after the Oct 30 Preview. About 9B parameters (about 14.6B inference-equivalent): a shared block of 24 decoder layers runs twice, giving 56 layer executions from 32 distinct layers, with two-stream manifold-constrained hyper-connections and hybrid linear/full attention (3:1). Text and image input, text output, thinking mode, MTP; context up to 512K tokens (524,288 via YaRN factor 2 over 262,144). Official AWQ 4-bit and GGUF builds; SenseTime says 256K-context multimodal inference fits in about 15 GiB with 4-bit weights and 8-bit KV cache. Reports AutomationBench 36.67 and GAIA 69.90, up from 14.33 and 37.86 for the Preview (vendor-reported).

Specifications

License
Open source · Apache-2.0
Weights
Downloadable
Architecture
hybrid
Parameters
9B
Context window
524K tokens
Max output
—
Knowledge cutoff
—
Price (in / out, $/M)
—
Modalities
TextVisionCode

Benchmarks

No benchmark scores recorded yet. Spotted some? Submit a correction.

Vendor-reported figures are claims until independently verified. See methodology.