🤗 Hugging Face Trending Models 2026-04-06

[github-actions]

Hugging Face Trending Models Digest 2026-04-06

Source: Hugging Face Hub | 30 models | Generated: 2026-04-06 00:11 UTC

---

Hugging Face Trending Models Digest — April 6, 2026

1. Today's Highlights

Google's Gemma 4 family dominates this week's leaderboard with multiple variants spanning 2B to 31B parameters, including experimental "any-to-any" multimodal models (E2B/E4B). Community distillation efforts are accelerating, with Jackrong's Qwen3.5-Claude-4.6-Opus reasoning distillates capturing significant attention—demonstrating demand for compact, high-reasoning models. The 1-bit quantization movement gains traction through prism-ml's Bonsai-8B, signaling a push toward extreme edge deployment. Notably, uncensored fine-tunes from HauhauCS are driving substantial download volume, reflecting persistent market demand for unaligned model variants. Netflix's entry into video-to-video generation with void-model hints at expanding enterprise AI investments beyond traditional tech players.

---

2. Trending Models by Category

🧠 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	2,345	539,356	Community-distilled reasoning model capturing Claude-4.6-Opus capabilities in a 27B parameter efficient architecture.
HauhauCS/Qwen3.5-9B-Uncensored-HauhauCS-Aggressive	HauhauCS	994	725,110	High-engagement uncensored variant with exceptional download velocity, indicating strong demand for unfiltered local inference.
LiquidAI/LFM2.5-350M	LiquidAI	229	14,877	Ultra-compact 350M parameter liquid neural network for resource-constrained applications.
chromadb/context-1	chromadb	370	3,584	Purpose-built retrieval-augmented generation model from the vector database leader.
facebook/tribev2	facebook	307	53,322	Meta's latest community-oriented model with regional US licensing considerations.

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

Model	Author	Likes	Downloads	Why It's Trending
google/gemma-4-31B-it	google	983	490,192	Flagship instruction-tuned multimodal model with native vision-language capabilities.
google/gemma-4-E4B-it	google	349	197,704	Experimental "any-to-any" architecture enabling flexible input-output modality combinations.
google/gemma-4-E2B-it	google	250	167,166	Compact any-to-any variant demonstrating Google's modality-agnostic model direction.
baidu/Qianfan-OCR	baidu	987	37,707	Production-grade Chinese-optimized document understanding with InternVL architecture.
netflix/void-model	netflix	389	0	Enterprise video inpainting and object removal—first major streaming platform open-weight release.
CohereLabs/cohere-transcribe-03-2026	CohereLabs	804	120,998	Cohere's latest ASR model with competitive performance on long-form audio.
k2-fsa/OmniVoice	k2-fsa	244	39,852	Zero-shot multilingual voice cloning with next-token prediction architecture.
mistralai/Voxtral-4B-TTS-2603	mistralai	673	5,458	Mistral's compact text-to-speech entry optimized for vLLM deployment.
Hcompany/Holo3-35B-A3B	Hcompany	222	1,532	MoE-based multimodal model with aggressive activation sparsity (35B total, 3B active).
HauhauCS/Qwen3.5-35B-A3B-Uncensored-HauhauCS-Aggressive	HauhauCS	1,185	688,440	Uncensored MoE vision-language model combining high engagement with multimodal capabilities.

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

Model	Author	Likes	Downloads	Why It's Trending
microsoft/harrier-oss-v1-0.6b	microsoft	163	10,824	Qwen3-based embedding model optimized for MTEB benchmarks and sentence-transformers compatibility.

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

Model	Author	Likes	Downloads	Why It's Trending
prism-ml/Bonsai-8B-gguf	prism-ml	426	38,631	Extreme 1-bit quantization enabling sub-1GB deployment of 8B parameter models.
Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-v2-GGUF	Jackrong	509	254,428	Refined GGUF conversion of the popular reasoning distillation with improved llama.cpp compatibility.
unsloth/gemma-4-26B-A4B-it-GGUF	unsloth	245	487,384	Unsloth-optimized GGUF conversion with 4-bit activation-aware quantization.
unsloth/gemma-4-31B-it-GGUF	unsloth	179	408,920	High-download conversion of Google's flagship multimodal model for local inference.
dealignai/Gemma-4-31B-JANG_4M-CRACK	dealignai	201	3,123	MLX-native abliterated variant for Apple Silicon deployment with uncensored behavior.
nvidia/Gemma-4-31B-IT-NVFP4	nvidia	195	85,892	NVIDIA-optimized FP4 quantization using ModelOpt for Hopper/Blackwell efficiency.
Jackrong/Qwopus3.5-9B-v3-GGUF	Jackrong	187	43,478	Compact reasoning-optimized Qwen variant with multi-platform GGUF support.
Jackrong/Qwopus3.5-27B-v3-GGUF	Jackrong	171	37,479	Larger reasoning-focused variant with explicit reasoning tag in model metadata.
prism-ml/Bonsai-8B-mlx-1bit	prism-ml	147	20,522	Apple Silicon-optimized 1-bit implementation with MLX-Swift compatibility.
unsloth/gemma-4-E4B-it-GGUF	unsloth	140	317,335	Any-to-any model conversion enabling local multimodal experimentation.

---

3. Ecosystem Signal

Gemma 4 has emerged as the definitive open-weight release of early 2026, with Google deploying a sophisticated tiered strategy: base multimodal models (31B/26B), activation-sparse variants (A4B), and experimental any-to-any architectures (E2B/E4B). This represents a maturation beyond simple scaling—Google is now competing on architectural diversity and deployment flexibility. The Qwen3.5 ecosystem, particularly through community distillation efforts by Jackrong and HauhauCS, demonstrates that Chinese-origin model families have achieved global adoption parity, with reasoning distillation and uncensored variants driving substantial engagement.

Quantization activity is intensifying along two axes: extreme compression (1-bit Bonsai, NVFP4) and ecosystem-native optimization (MLX for Apple Silicon, GGUF for llama.cpp). Unsloth has established itself as the dominant conversion infrastructure, appearing across four high-download Gemma variants. The uncensored model segment, while controversial, shows no demand abatement—HauhauCS variants account for over 1.4M combined downloads. Enterprise participation is expanding meaningfully: Netflix's video-to-video release and Baidu's OCR specialization indicate open-weight strategies extending beyond traditional AI labs into vertical application domains.

---

4. Worth Exploring

Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled — This represents the most sophisticated open-weight reasoning distillation available, compressing frontier Claude capabilities into a deployable 27B parameter footprint. The 2,345 likes and half-million downloads suggest validated quality; worth studying for distillation methodology and as a production reasoning backend.

prism-ml/Bonsai-8B-gguf — The 1-bit quantization frontier merits close attention for edge AI implications. With 38K+ downloads despite technical novelty, Bonsai demonstrates practical viability of extreme quantization. Essential for researchers and engineers evaluating sub-1GB model deployment.

google/gemma-4-E4B-it — Google's "any-to-any" architecture is the most significant structural innovation in this release cycle. With 197K downloads for an experimental variant, it offers early access to modality-agnostic model paradigms that may define next-generation multimodal systems.

---

This digest is auto-generated by agents-radar.