SenseNova Loop-9B model released by Megvii
The story so far
Megvii Technology released the SenseNova Loop-9B model, which extends computational depth under 9B parameters through Loop structure and manifold constraints, with weights of about 4.9GB after 4-bit quantization and low deployment memory requirements of up to 10GB. The model's context length can be up to 512K tokens, showing competitiveness comparable to Qwen3.8-27B in Agent and data analysis evaluations: scoring 41.36 on NL2Repo, close to the latter's 44.60; scoring 65.31 on AIDABench, higher than the latter's 65.01. Megvii has opened the model weights and related code for developers and researchers to download, deploy, and explore.
AI-written from the reports · Updated 2 hours ago
Coverage timeline
Follow the coverage to see every side of the story.
- ITHome · AISenseNova Loop-9B model released by SenseTime: deployment memory requirement as low as 10 GB
SenseTime releases SenseNova Loop-9B model, extending computational depth under 9B parameters through Loop structure and manifold constraints, with 4-bit quantization weights of about 4.9GB, deployment memory requirement as low as 10GB. The model's context length is up to 512K tokens, achieving a score of 41.36 in NL2Repo and 65.31 in AIDABench in Agent and data analysis evaluations, comparable to Qwen3.8-27B.
Heat over time
Not enough data yet to draw a trend.