Skip to content

🤗 Hugging Face Trending Models 2026-05-08 #994

Description

@github-actions

Hugging Face Trending Models Digest 2026-05-08

Source: Hugging Face Hub | 30 models | Generated: 2026-05-08 00:20 UTC


Hugging Face Trending Models Digest — 2026-05-08

Today's Highlights

This week's trending landscape is dominated by the DeepSeek-V4 family and Qwen3.6 series, with both seeing massive download volumes and significant community fine-tuning activity. Google's Gemma-4-31B-it continues its strong run, crossing 8.5 million downloads, while Gemma-4-E4B-it (a smaller 4B variant) also surges past 5 million downloads, signaling strong adoption of Google's latest open-weight family. The emergence of uncensored and "abliterated" fine-tunes (e.g., dealignai/Gemma-4-31B-JANG_4M-CRACK, HauhauCS/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive) points to a sustained community interest in less-restricted model variants. On the multimodal side, Sulphur-2-base (text-to-video) and OmniVoice (zero-shot voice cloning) are notable newcomers gaining rapid traction.


Trending Models

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

  • deepseek-ai/DeepSeek-V4-Pro
    Author: deepseek-ai | Likes: 3,724 | Downloads: 946,264
    Flagship text-generation model from DeepSeek's V4 family; trending due to strong benchmarks and massive community adoption for conversational AI.

  • deepseek-ai/DeepSeek-V4-Flash
    Author: deepseek-ai | Likes: 980 | Downloads: 751,333
    Faster, lighter variant of DeepSeek-V4 optimized for inference speed; complements the Pro version for production use cases.

  • XiaomiMiMo/MiMo-V2.5-Pro
    Author: XiaomiMiMo | Likes: 469 | Downloads: 20,905
    Long-context text-generation model from Xiaomi, with agent capabilities; gaining interest for enterprise and agentic workflows.

  • poolside/Laguna-XS.2
    Author: poolside | Likes: 232 | Downloads: 16,792
    Text-generation model for code and software engineering; trending as poolside expands its developer-focused model lineup.

  • ibm-granite/granite-4.1-8b
    Author: ibm-granite | Likes: 162 | Downloads: 24,099
    IBM's latest 8B Granite model for text generation; notable for enterprise-grade licensing and language coverage.

  • talkie-lm/talkie-1930-13b-it
    Author: talkie-lm | Likes: 242 | Downloads: 0
    13B instruction-tuned chat model; listed but with zero downloads—likely recently uploaded or under review.

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

  • google/gemma-4-31B-it
    Author: google | Likes: 2,555 | Downloads: 8,594,149
    Google's flagship multimodal model (image-text-to-text); the most downloaded model this week, reflecting strong open-weight competition.

  • google/gemma-4-E4B-it
    Author: google | Likes: 942 | Downloads: 5,494,056
    Smaller 4B parameter Gemma-4 variant (any-to-any); highly downloaded, ideal for efficient deployment on consumer hardware.

  • Qwen/Qwen3.6-35B-A3B
    Author: Qwen | Likes: 1,659 | Downloads: 3,211,156
    Mixture-of-experts multimodal model (35B total, 3B active); trending for high performance-to-compute ratio and vision-language tasks.

  • Qwen/Qwen3.6-27B
    Author: Qwen | Likes: 1,174 | Downloads: 1,771,851
    Dense multimodal model from Qwen's 3.6 family; strong adoption for image-text-to-text applications.

  • google/gemma-4-31B-it-assistant
    Author: google | Likes: 139 | Downloads: 19,908
    Assistant-optimized variant of Gemma-4-31B with any-to-any pipeline; targets conversational agents and tool-use scenarios.

  • google/gemma-4-26B-A4B-it-assistant
    Author: google | Likes: 84 | Downloads: 12,868
    MoE assistant variant (26B total, 4B active); efficient any-to-any model for multimodal assistant use cases.

  • sensenova/SenseNova-U1-8B-MoT
    Author: sensenova | Likes: 187 | Downloads: 2,724
    Multimodal any-to-any model (8B) with feature extraction; from SenseTime's SenseNova lineup.

  • nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16
    Author: nvidia | Likes: 262 | Downloads: 65,066
    NVIDIA's reasoning-optimized multimodal MoE (30B total, 3B active); notable for combining any-to-any capabilities with reasoning.

  • Zyphra/ZAYA1-8B
    Author: Zyphra | Likes: 194 | Downloads: 539
    8B model with Apache-2.0 license and eval results; lightweight multimodal model with permissive licensing.

  • mistralai/Mistral-Medium-3.5-128B
    Author: mistralai | Likes: 295 | Downloads: 18,272
    Mistral's large 128B model for multilingual (EN/FR) text generation; vLLM-compatible.

  • SulphurAI/Sulphur-2-base
    Author: SulphurAI | Likes: 375 | Downloads: 71,149
    Text-to-video model with Diffusers and GGUF support; a major new entrant in the video generation space.

  • TenStrip/LTX2.3-10Eros
    Author: TenStrip | Likes: 150 | Downloads: 28,215
    Image-to-video generation model; trending for creative video synthesis applications.

  • SeeSee21/Z-Anime
    Author: SeeSee21 | Likes: 216 | Downloads: 4,460
    Text-to-image anime-focused model with Diffusers and GGUF support; popular in the anime generation community.

  • k2-fsa/OmniVoice
    Author: k2-fsa | Likes: 797 | Downloads: 2,238,817
    Zero-shot, multilingual text-to-speech with voice cloning; surging in popularity for voice AI applications.

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

  • openai/privacy-filter
    Author: openai | Likes: 1,343 | Downloads: 165,240
    Token-classification model for privacy filtering (ONNX, safetensors); trending as data privacy tools gain urgency.

  • froggeric/Qwen-Fixed-Chat-Templates
    Author: froggeric | Likes: 92 | Downloads: 0
    Utility model providing corrected chat templates for Qwen models; useful for developers fixing template issues.

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

  • dealignai/Gemma-4-31B-JANG_4M-CRACK
    Author: dealignai | Likes: 1,487 | Downloads: 169,511
    "Abliterated" (uncensored) fine-tune of Gemma-4-31B; highly controversial but very popular for unrestricted applications.

  • unsloth/Qwen3.6-35B-A3B-GGUF
    Author: unsloth | Likes: 956 | Downloads: 2,417,319
    GGUF quantization of Qwen3.6-35B-A3B; the most downloaded GGUF this week, essential for local inference.

  • unsloth/Qwen3.6-27B-GGUF
    Author: unsloth | Likes: 612 | Downloads: 1,264,430
    GGUF version of Qwen3.6-27B; widely used for efficient CPU/edge deployment.

  • HauhauCS/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive
    Author: HauhauCS | Likes: 576 | Downloads: 973,262
    Uncensored, MoE vision-language fine-tune in GGUF; trending for aggressive uncensoring of Qwen3.6.

  • z-lab/Qwen3.6-27B-DFlash
    Author: z-lab | Likes: 259 | Downloads: 28,767
    Specialized fine-tune of Qwen3.6-27B with DFlash; targets enhanced feature extraction and generation performance.

  • DavidAU/Qwen3.6-27B-Heretic-Uncensored-FINETUNE-NEO-CODE-Di-IMatrix-MAX-GGUF
    Author: DavidAU | Likes: 91 | Downloads: 126,660
    Uncensored code-oriented fine-tune of Qwen3.6-27B in GGUF format; combines uncensoring with code specialization.

  • Jackrong/Qwopus3.6-35B-A3B-v1-GGUF
    Author: Jackrong | Likes: 68 | Downloads: 4,754
    GGUF quant of a Qwen3.6-35B variant; part of the growing Qwen3.6 quantization ecosystem.

  • Jackrong/Qwen3.5-9B-DeepSeek-V4-Flash-GGUF
    Author: Jackrong | Likes: 102 | Downloads: 104,723
    Hybrid GGUF merging Qwen3.5-9B with DeepSeek-V4-Flash capabilities; creative community merge.


Ecosystem Signal

The current trend landscape reveals several key dynamics. DeepSeek-V4 and Qwen3.6 families are the dominant "superclusters" this week, with combined download volumes in the millions and extensive community fine-tuning activity. Google's Gemma-4 series continues to gain enterprise traction, particularly the 31B and 4B variants, while the Gemma-4-E4B-it (4B) variant is notably one of the fastest-growing models, suggesting strong demand for efficient multimodal models.

Uncensored/abliterated fine-tunes (Gemma-4-JANG, Qwen3.6-Heretic, Qwen3.6-Uncensored-Aggressive) are a persistent ecosystem trend—these models consistently attract high likes and downloads, indicating a vocal community segment prioritizing unrestricted outputs despite potential safety concerns. GGUF quantization remains the dominant format for local deployment, with unsloth leading GGUF conversions. The text-to-video space is heating up with Sulphur-2-base, while OmniVoice signals growing interest in zero-shot voice cloning. Open-weight models continue to dominate vs proprietary—every trending model above is open-weight or has open fine-tunes.


Worth Exploring

  1. google/gemma-4-E4B-it — At just 4B parameters with any-to-any multimodal capabilities, this is an ideal model for researchers and developers seeking a lightweight yet capable multimodal model for edge deployment or rapid prototyping. Its 5.5M downloads confirm strong practical utility.

  2. k2-fsa/OmniVoice — With 2.2M downloads and 797 likes, this zero-shot multilingual TTS model is a standout in the audio generation space. Worth studying for its voice cloning approach and potential applications in accessibility, content creation, and voice assistants.

  3. dealignai/Gemma-4-31B-JANG_4M-CRACK — Whether for research on model safety, alignment, or the ethics of uncensored models, this abliterated fine-tune of Gemma-4-31B is a key artifact for understanding community-driven efforts to bypass safety guardrails. Its high engagement (1,487 likes) makes it culturally significant in the open-weight ecosystem.


This digest is auto-generated by agents-radar.

Metadata

Metadata

Assignees

No one assigned

    Labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions