Open-source AI weekly
This week
- 3 of 15
- new models on the list
- 7 of 15
- under a permissive licence
- 6 of 12
- tracked projects released
- 130 of 206
- listed papers with public code
Models
Downloads cover the last 30 days. Custom means the maker wrote its own licence terms.
Language and vision
| Model | Licence | Likes | Downloads |
|---|---|---|---|
| Kimi-K3.1-Pro New moonshotai · Text and images | Custom | 1,691 | 25,569 |
| Qwen3.8-27B Qwen · Text and images | Apache 2.0 | 19,373 | 7,668,187 |
| Arete-27B New NousResearch · Text | Apache 2.0 | 259 | 1,747 |
| Qwen3.5-4B-GLB-3D-Coder New ccore · Text | Apache 2.0 | 267 | 33,641 |
| DeepSeek-V4.1-Flash-1017 deepseek-ai · Text and images | MIT | 5,405 | 295,747 |
More on this area in Foundation Models.
Speech
| Model | Licence | Likes | Downloads |
|---|---|---|---|
| NVIDIA-Nemotron-3.5-Transcribe-30B-A4B nvidia · Transcription | Custom | 3,104 | 67,230 |
| Audio8-TTS-Realistic-Air Edge0 · Voice | Apache 2.0 | 358 | 15,319 |
| drama-3 fishaudio · Voice | Custom | 1,226 | 43,735 |
| Breeze-TTS-2 BreezeBlue · Voice | Custom | 953 | 25,640 |
| Ming-3.0-tts-0.5B inclusionAI · Voice | MIT | 217 | 12,883 |
More on this area in Voice & Speech AI.
Image and video
| Model | Licence | Likes | Downloads |
|---|---|---|---|
| LTX-2.5 Lightricks · Image to video | Custom | 11,065 | 2,661,026 |
| Pika-H3-Flex Pika-Labs · Image to video | Custom | 514 | 1,083 |
| P-H3-Image PrunaAI · Image | Custom | 166 | 893 |
| MOSS-Image-SFT OpenMOSS-Team · Image | Apache 2.0 | 934 | 20,971 |
| Qwen-Image-2.1 Qwen · Image | Custom | 3,585 | 251,760 |
More on this area in Avatars & Vision.
Releases
Stable versions published this week by the open projects we follow.
| Project | Version | Date |
|---|---|---|
| transformers | v5.22.2 | 6 November 2026 |
| agents | livekit-agents@1.8.11 | 5 November 2026 |
| ComfyUI | v0.44.0 | 4 November 2026 |
| ollama | v0.40.15 | 3 November 2026 |
| open-webui | v0.13.0 | 3 November 2026 |
| vllm | v0.33.0rc3 | 2 November 2026 |
Papers with code
The most upvoted papers on Hugging Face's daily list this week that link to public code.
- Wave Forcing: Towards Speed-of-Light Streaming Video Generation 125 upvotes · code · 130 stars
- CoworkOS:Scaling Agent Coworkers for Real-world Automation 100 upvotes · code · 142 stars
- Route Anything: A Study of Model and Policy Routing for General-Purpose AI Systems 80 upvotes · code · 18 stars
- Video Generation Meets Agent Harness: Agentic Video Generation, Agentic World Modeling and Recursive Self-Improvement 78 upvotes · code · 57 stars
- Words Caught in the Heart: Simulating Person-Specific Expression for Self-Opaque Clients in Psychotherapy 77 upvotes · code · 1 stars
- MoviEdit: Long-form Multi-Shot Video Editing Made Easy 74 upvotes · code · 18 stars
- Harness-S: Learning from Sibling Harnesses for Generalizable Agent Training 66 upvotes · code · 58 stars
- Think Less, Locally: Local Preference Learning for Reasoning Compression 61 upvotes · code · 11 stars
How this page works
Models come from Hugging Face's trending lists for the tasks a digital human depends on. We keep models under 120 days old with at least 100 likes, take one entry per maker, and leave out compressed copies, add-ons and adult models. Releases come from the projects' GitHub pages and papers from Hugging Face's daily list, with records on arXiv.
The numbers are public counters, read once a week. We have not run these models, and a place on the list is not a recommendation. Earlier weeks stay on this page below.
Archive
Week of 25 October 2026 to 31 October 2026
Open weights do not mean an open licence. Eight of the 15 models on this week's list ship under their maker's own terms, among them LTX-2.5, Pika-H3-Flex and Mistral-Large-4. If you plan to put one in a product, read the terms before you benchmark it.
Speech is the lane to watch if you are building a digital human. Four of its five entries are voice models, and two of those, Audio8-TTS-Realistic and maya2-native, are under Apache 2.0.
Likes and downloads tell different stories. Pika-H3-Flex collected about 300 likes in three days and a few hundred downloads, while Unlimited-OCR-2 passed 100,000 downloads in under a week. Read likes as interest and downloads as use.
| Model | Licence | Downloads |
|---|---|---|
| DeepSeek-V4.1-Flash-1017 deepseek-ai · Text and images | MIT | 170,353 |
| LFM3-1.7B LiquidAI · Text | Custom | 37,568 |
| Nanbeige4.3-33B Nanbeige · Text | Apache 2.0 | 15,795 |
| Mistral-Large-4-1T-A52B mistralai · Text and images | Custom | 4,576 |
| Unlimited-OCR-2 baidu · Text and images | MIT | 109,519 |
| NVIDIA-Nemotron-3.5-Transcribe-30B-A4B nvidia · Transcription | Custom | 29,364 |
| Audio8-TTS-Realistic-0.6b Edge0 · Voice | Apache 2.0 | 4,613 |
| maya2-native maya-research · Voice | Apache 2.0 | 3,152 |
| drama-3 fishaudio · Voice | Custom | 33,436 |
| Breeze-TTS-2 BreezeBlue · Voice | Custom | 21,483 |
| LTX-2.5 Lightricks · Image to video | Custom | 2,431,846 |
| Pika-H3-Flex Pika-Labs · Image to video | Custom | 404 |
| MOSS-Image-SFT OpenMOSS-Team · Image | Apache 2.0 | 16,757 |
| Qwen-Image-2.1 Qwen · Image | Custom | 217,429 |
| Minimax-h3_Singularity WarmBloodAban · Image to video | Apache 2.0 | 986,743 |
Releases: transformers v5.22.1, open-webui v0.12.5, pipecat v1.15.0, agents livekit-agents@1.8.10, ollama v0.40.13, diffusers v0.42.1, ComfyUI v0.43.0.