🤗 Hugging Face Trending Models 2026-04-04

[github-actions]

Hugging Face Trending Models Digest 2026-04-04

Source: Hugging Face Hub | 30 models | Generated: 2026-04-04 00:10 UTC

---

Hugging Face Trending Models Digest — April 4, 2026

---

1. Today's Highlights

The Qwen 3.5 ecosystem dominates this week's trending charts, with community fine-tuner Jackrong claiming the top spot through aggressive distillation of Claude 4.6 Opus reasoning capabilities into a 27B parameter model. Google's Gemma-4 family has launched in force with multiple variants spanning 2B to 31B parameters, including experimental "any-to-any" multimodal architectures. Notably, uncensored fine-tunes from HauhauCS are driving massive download volumes—nearly 1.4M combined—suggesting strong demand for unfiltered local inference. The emergence of 1-bit quantization (prism-ml's Bonsai) and Netflix's entry into video AI signal continued diversification beyond pure language modeling.

---

2. Trending Models

🧠 Language Models (LLMs, chat models, instruction-tuned)

Model	Author	Likes	Downloads	Why It's Trending
Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled	Jackrong	2,222	487,446	Distills Claude 4.6 Opus reasoning into open-weights Qwen 3.5, offering frontier-level chain-of-thought at runnable scale
google/gemma-4-31B-it	google	674	76,200	Google's flagship instruction-tuned Gemma-4 with native vision-language capabilities
nvidia/Nemotron-Cascade-2-30B-A3B	nvidia	456	137,849	NVIDIA's Cascade MoE architecture delivering efficient inference through activated parameter sparsity
LiquidAI/LFM2.5-350M	LiquidAI	212	10,194	Ultra-efficient 350M parameter liquid foundation model for edge deployment
chromadb/context-1	chromadb	362	3,195	Chroma's first native generative model, optimized for retrieval-augmented generation workflows

🎨 Multimodal & Generation (image, video, audio, text-to-X)

Model	Author	Likes	Downloads	Why It's Trending
Qwen/Qwen3.5-9B	Qwen	1,157	4,818,944	Official Qwen 3.5 release with 4.8M downloads—dominant open vision-language foundation
baidu/Qianfan-OCR	baidu	859	26,980	Baidu's production OCR with InternVL architecture, strong CJK document understanding
google/gemma-4-E4B-it	google	253	23,460	Experimental "any-to-any" modality—accepts and generates arbitrary media types
mistralai/Voxtral-4B-TTS-2603	mistralai	649	4,760	Mistral's compact multilingual TTS with vLLM inference optimization
k2-fsa/OmniVoice	k2-fsa	128	6,560	Open-source TTS supporting African American English and African American Language variants
netflix/void-model	netflix	197	0	Netflix's video inpainting model for object removal—zero downloads suggest gated release
facebook/sam3.1	facebook	126	7,763	Segment Anything 3.1 with native video segmentation capabilities

🔧 Specialized Models (code, math, medical, embeddings)

Model	Author	Likes	Downloads	Why It's Trending
CohereLabs/cohere-transcribe-03-2026	CohereLabs	763	84,600	Cohere's latest ASR with competitive Whisper-level accuracy and enterprise licensing
microsoft/harrier-oss-v1-0.6b	microsoft	148	7,508	600M parameter embedding model built on Qwen3, MTEB-benchmarked
microsoft/harrier-oss-v1-270m	microsoft	117	5,677	Lightweight 270M embedding variant using Gemma3 text encoder

📦 Fine-tunes & Quantizations (community fine-tunes, GGUF, AWQ)

Model	Author	Likes	Downloads	Why It's Trending
HauhauCS/Qwen3.5-9B-Uncensored-HauhauCS-Aggressive	HauhauCS	948	700,218	Highest-volume uncensored fine-tune—"aggressive" safety removal for unrestricted local use
HauhauCS/Qwen3.5-35B-A3B-Uncensored-HauhauCS-Aggressive	HauhauCS	1,156	652,331	MoE-scale uncensored variant with vision capabilities, nearly matching official Qwen's like count
Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-v2-GGUF	Jackrong	492	227,053	GGUF quantization of the top-trending reasoning model—broad llama.cpp compatibility
prism-ml/Bonsai-8B-gguf	prism-ml	357	26,164	Extreme 1-bit quantization—8B model runnable on consumer hardware with minimal quality loss
prism-ml/Bonsai-8B-mlx-1bit	prism-ml	126	12,610	Apple Silicon-optimized MLX port of the 1-bit Bonsai architecture
unsloth/gemma-4-26B-A4B-it-GGUF	unsloth	182	99,223	Unsloth-optimized GGUF of Google's MoE Gemma-4 with 4B active parameters
unsloth/gemma-4-31B-it-GGUF	unsloth	130	83,931	Dense Gemma-4 31B in efficient GGUF format for local inference
Jackrong/Qwopus3.5-9B-v3-GGUF	Jackrong	146	19,991	Compact reasoning-optimized Qwen 3.5 variant in GGUF
Jackrong/Qwopus3.5-27B-v3-GGUF	Jackrong	116	10,567	Larger-scale reasoning variant with aggressive quantization for accessibility

---

3. Ecosystem Signal

Qwen 3.5 has achieved ecosystem dominance—the official 9B base model leads all downloads at 4.8M, while community fine-tunes from Jackrong and HauhauCS capture three of the top five like counts. This mirrors Llama's 2024 trajectory: Alibaba's permissive licensing and consistent quality releases have cultivated a thriving derivative market. Google's Gemma-4 launch represents a strategic pivot—abandoning the "small models only" positioning for competitive 31B dense and MoE variants, though download volumes trail Qwen significantly.

The uncensored fine-tuning economy has matured into a measurable force. HauhauCS's combined 1.35M downloads demonstrate that safety-filter removal drives substantial user demand, particularly for local deployment. This creates tension with platform policies and model licenses, yet remains technically straightforward with current tooling.

Quantization innovation is accelerating: 1-bit methods (Bonsai) and Unsloth's optimized GGUF pipelines are collapsing hardware requirements faster than model growth. The ratio of GGUF downloads to base model downloads (~50% for top models) suggests local inference is now default behavior for power users. Meanwhile, proprietary frontier signals appear through distillation—Jackrong's "Claude 4.6 Opus" branding implies access to unreleased Anthropic capabilities, highlighting the continued information asymmetry between closed and open development.

---

4. Worth Exploring

Model	Recommendation
Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled	Most significant capability unlock. If the distillation claim holds, this offers frontier reasoning at 27B scale—test against GPQA or live coding benchmarks. The 487K downloads suggest community validation, but verify reasoning traces independently.
prism-ml/Bonsai-8B-gguf	Technical frontier to watch. 1-bit quantization has been theoretically promising but practically disappointing; Bonsai's 26K downloads and positive early signals warrant testing on your own hardware. Success here would redefine edge AI economics.
google/gemma-4-E4B-it	Architectural preview. The "any-to-any" designation suggests Google's unified multimodal ambitions. With only 23K downloads, it's underexplored relative to hype—probe for actual cross-modal capabilities beyond marketing, particularly audio and video generation quality.

---

---

This digest is auto-generated by agents-radar.