Models · daily board

What's moving on the Hub

5,276 models with real traffic were created in the last seven days. What is trending, what is being pulled faster than it was, which base models the field is building on, what arrived this week and what everyone already uses — each with its size, licence and lineage, read nightly from every model that has traffic.

202,261 models with traffic · read 2026-09-17 · 7 days of download and like history since 2026-09-10

New models per day

2026-07-1937,007 over 60 · peak 1,168 on 2026-08-182026-09-16

Models that reached the traffic tier, by the day they were created. A spike is a release; the plateau is the field's daily rate.

#	Model	Trend score	Downloads 30d	Likes	Params	Commercial
1	deepseek-ai/DeepSeek-V4.1-Flash image-text-to-text	2,406	366k	2,784	763B	yes
2	Edge0/Edge0-35B-A3B-preview text-generation	2,268	28k	2,984	35B	yes
3	openbmb/MiniCPM5-2B text-generation	618	324k	1,490	2.5B	yes
4	Qwen/Qwen3.8-27B image-text-to-text	582	7.7M	15,335	28B	yes
5	m-a-p/YuE2-3B text-to-audio	566	9,391	600	3.6B	no
6	nex-agi/Nex-N2.5-mini text-generation	485	6,837	817	35B	yes
7	ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF image-text-to-text	431	957k	1,163	28B base model	yes
8	Lightricks/LTX-2.5 image-to-video	400	1.6M	4,055	—	unknown
9	TokenRhythm/NeoHorse-1-4B text-generation	394	16k	2,094	4.2B	yes
10	unsloth/Qwen3.8-27B-GGUF	337	8.9M	4,188	28B base model	yes
11	DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NEO-CODER-MAX-MTP-GGUF image-text-to-text	330	1.0M	764	28B base model	yes
12	ukisai/Swift-Qwen3.8-27b image-text-to-text	291	2,753	297	28B	unknown
13	sentence-transformers/all-MiniLM-L6-v2 sentence-similarity	251	256M	6,017	23M	yes
14	tencent/AuK text-to-speech	244	2,753	264	—	yes
15	XHToken/Spark-X2.5-4B text-generation	229	27k	1,223	4.1B	yes
16	WarmBloodAban/Minimax-h3_Singularity image-to-video	226	164k	439	33B base model	yes
17	MiniMaxAI/MiniMax-H3 image-text-to-video	215	4.7M	5,363	33B	unknown
18	openai-community/gpt2 text-generation	209	16M	4,112	137M	yes
19	Qwen/Qwen3.8-Flash-Next image-text-to-text	205	689k	5,292	180B	unknown
20	meta-llama/Llama-3.1-8B-Instruct text-generation	202	5.9M	7,648	8.0B	conditional

Trend score is Hugging Face's own, read live. Downloads are the Hub's rolling thirty-day count. The licence column is the resolved verdict, not the card's string.

How the corpus is made

unstated67k+141%

text-generation66k+126%

text-to-image16k+19.9%

image-text-to-text14k+26.4%

sentence-similarity7,091+12.8%

automatic-speech-recognition3,932+47.5%

text-classification3,675+316%

text-to-speech3,020+69.3%

feature-extraction2,523+173%

token-classification2,196+161%

image-classification1,875+182%

robotics1,450+503%

translation1,170+60.2%

image-to-image1,159-19.0%

text-to-video983+8.4%

fill-mask930+76.5%

0share of 193,330 models · pill: last 30 days vs the 30 before67k

Lifetime share on the bar; the pill is the last thirty days of arrivals against the thirty before, so a category growing faster than the corpus is where the new models are going. Size classes come from parameter counts; "unstated" is a model whose card and weights give none.

Which of these you can run under which runtime, and under what terms, is worked through in Which models can you actually run? and What you're allowed to ship.