This is more than in all of 2025. MDASH is Microsoft Security's multi-model agentic scanning harness that now scans Windows code at greater scale.
Microsoft Source: Microsoft news post, published Nov 20, 2026
Qwen's models drew 188 million downloads on Hugging Face in 30 days, 8 times the next organization.
Updates: live snapshot · Last checked Nov 21, 2026
| Organization | Downloads, 30 days | Models counted | Most downloaded |
|---|---|---|---|
| Qwen (Alibaba) | 187,723,088 | 191 | Qwen3-0.6B |
| Microsoft | 24,364,312 | 131 | Florence-2-base |
| DeepSeek | 23,789,692 | 80 | DeepSeek-V3.2 |
| NVIDIA | 20,360,626 | 158 | Qwen3.6-35B-A3B-NVFP4 |
| 17,104,163 | 200 | gemma-4-31B-it | |
| Z.ai | 13,118,363 | 89 | GLM-5.3-Flash |
| Meta Llama | 10,185,867 | 62 | Llama-3.1-8B-Instruct |
| OpenAI | 8,015,464 | 5 | gpt-oss-20b |
| Hugging Face | 6,125,182 | 38 | SmolLM2-135M |
| Moonshot AI | 2,165,382 | 17 | Kimi-K3 |
| IBM Granite | 2,022,913 | 107 | granite-docling-258M |
| Mistral AI | 1,595,187 | 7 | Mistral-7B-Instruct-v0.2 |
Worked out from the chart's data, as of Nov 21, 2026.
Hugging Face reports how many times each model's files were downloaded in the last 30 days. We add up the language models of each organization, up to 100 models per type for text generation and image-and-text models.
Downloads count file requests, including automated ones, not unique users. The data refreshes daily from the Hugging Face Hub API.
On the chart: Downloads count file requests, including automated ones, not unique users. Up to 100 models per type per organization, text-generation and image-text-to-text.
Download as CSV · Free to use with credit to Interconnecting Intelligence and the original source.
| Organization | Downloads, 30 days | Models counted | Most downloaded |
|---|---|---|---|
| Qwen (Alibaba) | 187,723,088 | 191 | Qwen3-0.6B |
| Microsoft | 24,364,312 | 131 | Florence-2-base |
| DeepSeek | 23,789,692 | 80 | DeepSeek-V3.2 |
| NVIDIA | 20,360,626 | 158 | Qwen3.6-35B-A3B-NVFP4 |
| 17,104,163 | 200 | gemma-4-31B-it | |
| Z.ai | 13,118,363 | 89 | GLM-5.3-Flash |
| Meta Llama | 10,185,867 | 62 | Llama-3.1-8B-Instruct |
| OpenAI | 8,015,464 | 5 | gpt-oss-20b |
| Hugging Face | 6,125,182 | 38 | SmolLM2-135M |
| Moonshot AI | 2,165,382 | 17 | Kimi-K3 |
| IBM Granite | 2,022,913 | 107 | granite-docling-258M |
| Mistral AI | 1,595,187 | 7 | Mistral-7B-Instruct-v0.2 |
This is more than in all of 2025. MDASH is Microsoft Security's multi-model agentic scanning harness that now scans Windows code at greater scale.
Microsoft Source: Microsoft news post, published Nov 20, 2026
Microsoft is building containment, identity and manageability into Windows to run and manage AI agents securely. MXC policies can be set with Intune to control what agents can access.
MetaMicrosoftOpenAI Source: Microsoft news post, published Nov 20, 2026
The team plans to demonstrate a robot that can help a person eat and is experimenting with letting users give instructions in their own words.
University of Pittsburgh Source: Meta news post, published Nov 20, 2026
ATLAS is built on Lean, an open source language for writing mathematics that lets computers verify each logical step. It was presented at the summit as a shared foundation for researchers.
The project was launched with the Rothschild Hospital Foundation and the University of Montreal. It funds researchers to collect de-identified brain data and share it openly.
MetaRothschild Hospital FoundationUniversity of Montreal Source: Meta news post, published Nov 20, 2026