The Open Weights
LatestModelsLeaderboardsCompanies
Subscribe
The Open Weights

The daily record of open-source AI. New model releases, leaderboards, and what's coming next — written for people who ship.

Refreshed every 12 hours

Discover

  • Latest releases
  • New today
  • Trending models

Browse

  • All models
  • Companies
  • Categories
  • Leaderboards

About

  • About
  • Editorial policy
  • Hardware estimates
  • RSS feed
  • llms.txt
  • Newsletter

© 2026 The Open Weights. An independent publication.

PrivacyTermsSMSAggregated by Claude · curated by humans.
LatestSenseTimeLoop-9B
SenseTimeVision-Language

SenseTime Releases Looped 9B Multimodal Model

The company's new SenseNova-Loop-9B pairs vision-language capabilities with a recurrent looped-transformer design.

Nov 17, 2026
NotableOther
SenseNova-Loop-9B

SenseTime has published SenseNova-Loop-9B on Hugging Face, a 9-billion-parameter multimodal model that handles both image-and-text understanding and plain text generation. The release marks the debut of the company's Loop line, which leans on a looped-transformer design paired with what SenseTime calls an MHC architecture.

The looped-transformer approach is the notable idea here. Rather than stacking an ever-deeper tower of unique layers, looped designs reuse a smaller block of weights across repeated passes, effectively trading parameter count for recurrent depth at inference. That can let a comparatively modest model reason over more steps without ballooning its footprint.

Why it matters

  • At 9B parameters, the model sits in a size range that runs on a single workstation GPU, keeping it accessible to developers and researchers.
  • It is a vision-language model first, so it targets tasks that combine images and text rather than text alone.
  • The architecture is a departure from the standard dense-transformer recipe most open releases follow, giving the community a concrete artifact to test looped designs against.

The model ships under a custom license, so teams planning commercial use should read the terms on the repository before building on it. SenseTime has not published a context-length figure alongside the initial release, and independent benchmarks will be the real measure of whether the looped design pays off in practice.

Sources

  • sensenova/SenseNova-Loop-9B

    Hugging Face

    Visit
OlderTencent releases Hunyuan Hy4-flash multimodal MoETencent · Vision-Language · 3 days agoNewerReson8's Rumble-2-Flash Puts ASR in 0.5B ParamsReson8 · Speech → Text · 3 days ago

Get the model

Hugging Face

Specs

Parameters9B
Context window524K tokens
Size18.8 GB
PrecisionBF16
ArchitectureSenseNovaLoopForConditionalGeneration
LicenseOTHER
Downloads5.7K
Likes322

Can you run it?

Runs on a laptop — about 7.9 GB at 4-bit.

  • BF16 (as published)

    24 GB GPU (RTX 3090 / 4090) · 32 GB Mac

    21 GB
  • 8-bit

    16 GB GPU (RTX 4080 / 5070 Ti) · 24 GB Mac

    12.2 GB
  • 4-bit

    8 GB GPU (RTX 4060) · 16 GB laptop or Mac

    7.9 GB
Your machine
GGUF builds
  • KV cache approximated from model size (no config metadata) at 8,192 tokens.

Based on Hugging Face weights metadata. Assumes an 8K context and ~1 GB runtime overhead; actual needs vary. How we estimate


Modalities

Text / LLMVision-Language
2 versions — view changelog

The Weekly Weights

Every open release that mattered, one email a week.

0 comments

No comments yet. Be the first to weigh in.

More from SenseTime

All SenseTime releases →
SenseNova-Loop-9B
SenseTime/Vision-LanguageRuns on a laptop

SenseTime previews a 9B looped-transformer VLM

SenseNova-Loop-9B-Preview puts a recurrent-depth architecture to work on multimodal reasoning.

Oct 28, 2026
SenseTime/Any-to-AnyRuns on a laptop

SenseTime's SenseNova-U1.5 Unifies Vision Tasks

An 8B model drops the usual encoder and VAE in favor of a single native architecture spanning understanding, reasoning, and image generation.

Sep 9, 2026
SenseNova-U1.5-8B-MoT
SenseTime/Any-to-AnyRuns on a laptop

SenseTime Releases SenseNova U1.5 8B Any-to-Any Model

The new 8B multimodal model handles text, images, and image editing within a single native architecture.

Aug 19, 2026

More in Vision-Language

All Vision-Language →
Hunyuan Hy4-flash
Tencent/Vision-LanguageMulti-GPU server

Tencent releases Hunyuan Hy4-flash multimodal MoE

The lightweight, mixture-of-experts model adds vision and text instruction-following to Tencent's open Hunyuan lineup.

Nov 17, 2026
Mirage
Cactus Compute/Vision-Language

Cactus Compute's Mirage brings vision tasks on-device

The new open vision-language model targets detection, segmentation, and visual Q&A without relying on the cloud.

Nov 5, 2026
Inkling-2-Small
Thinkingmachines/Vision-LanguageMulti-GPU server

Thinking Machines ships Inkling-2-Small, an open MoE VLM

The Apache-2.0 vision-language model uses a mixture-of-experts design and lands openly on Hugging Face.

Nov 4, 2026