Technicals

Add as a preferred source on Google

Tencent model tops Hugging Face trending list with score of 316

Tencent's Hy4-flash topped Hugging Face's public trending ranking on Nov. 20 with a score of 316, while six models marked NEW entered the top 10.

Updated 2 min read
Hugging Face. (Photo: Shixart1985 / CC BY 2.0, cropped)
Hugging Face. (Photo: Shixart1985 / CC BY 2.0, cropped)

Tencent's Hy4-flash, an image-text-to-text model, topped Hugging Face's public trending ranking with a score of 316 on Nov. 20, while six newcomers entered the top 10, the ranking showed.

Higurashi-ai's Haru-4B ranked second with 309 points in the audio-to-audio category. Apple's FastFrames-L, a text-to-video model built on Wan-AI/Wan2.1-T2V-14B, ranked third with 292 and was marked NEW.

Inworld's voice-isolator ranked fourth with 265 for audio-to-audio, while harness-hub's Qwen3.8-27B-pi-adapted also scored 265 for image-text-to-text. The harness-hub model was marked NEW and built on Qwen/Qwen3.8-27B.

Qwen's Qwen3.8-27B ranked sixth with 238 points. MiniMaxAI's MiniMax-H3-Image ranked seventh with 198 for text-to-image and was built on MiniMaxAI/MiniMax-H3.

Lightricks' LTX-2.5 ranked eighth with 184 for image-to-video. Sensenova's SenseNova-Loop-9B ranked ninth with 162 for image-text-to-text and was marked NEW. Nvidia's NVIDIA-Nemotron-3.5-Bolt-E4B-4GB ranked 10th with 140 for text-generation and was built on nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-BF16.

Four other models followed: thinkingmachines/Inkling-2-Small scored 117, meituan-longcat/LongCat-Video-Avatar-1.6 scored 117, HuggingFaceTB/SmolVLM2-256M-Video-Instruct scored 116 and n4ze3m/tiny-voca scored 116. Unsloth's Qwen3.8-27B-GGUF scored 114.

The entries marked NEW were first uploaded within the previous 14 days. They were Hy4-flash on Nov. 17, Haru-4B on Nov. 12, FastFrames-L on Nov. 16, voice-isolator on Nov. 14, Qwen3.8-27B-pi-adapted on Nov. 16, SenseNova-Loop-9B on Nov. 17, LongCat-Video-Avatar-1.6 on Nov. 13 and tiny-voca on Nov. 14.

Qwen3.8-27B-pi-adapted and Qwen3.8-27B-GGUF are repackaged or adapted versions of Qwen3.8-27B. The list also includes other models built on earlier releases, including FastFrames-L, MiniMax-H3-Image, NVIDIA-Nemotron-3.5-Bolt-E4B-4GB and LongCat-Video-Avatar-1.6.

Qwen3.8-27B recorded 6.6 million downloads in the past 30 days, while Qwen3.8-27B-GGUF recorded 7.1 million. LTX-2.5 had 2.9 million downloads and MiniMax-H3-Image had 821,710, all on Hugging Face during the past 30 days.

Likes are counted across each model's lifetime. Qwen3.8-27B had 20,709 all-time likes, LTX-2.5 had 12,765 and Haru-4B had 3,681. Three models marketed as uncensored or similarly were left out.

The list was taken from Hugging Face's public trending ranking. Its score measures recent likes and activity on the site, rather than model quality or capability, while downloads cover the past 30 days and likes are all-time figures on Hugging Face.

Related articles

View all