Hugging Face Surging Models

공개

NeoDrop Official

Weekly digest of HF models with > 10x download growth, with brief description, license, and business applicability

Hugging Face Surging Models2026. 07. 07. 01:28:14

Qwythos tops a China-heavy HF week

This week’s Hugging Face breakout set is LLM-heavy: Qwythos has the clearest download spike, while Hy3, LongCat-2.0, and DeepSeek V4 DSpark are the strongest commercially usable large-model candidates. Non-LLM signals are mostly license-gated: Rampart is usable with attribution, while TabFM, OmniVoice, and Sulphur-2-base should stay in research or monitoring for now.

Hugging Face Surging Models2026. 06. 30. 01:33:49

HF Breakouts Jun 22–29: Agent Models Win

This week’s Hugging Face breakout set is strongest in agent infrastructure and document automation: Ornith-1.0, Qwen-AgentWorld, and Baidu Unlimited-OCR are the most commercially actionable picks. Krea-2 and LFM2.5-230M are useful with license constraints, while Sulphur-2-base and Higgs TTS 3 should stay out of commercial production until rights are clear.

Hugging Face Surging Models2026. 06. 23. 01:32:22

HF Breakout Models, Jun 15–22: GLM-5.2 and VibeThinker-3B

Two MIT-licensed LLM breakouts this week: GLM-5.2 (753B, ~41×, frontier agentic coding) and VibeThinker-3B (3B, ~32×, 96.1% LeetCode).

Hugging Face Surging Models2026. 06. 16. 01:27:39

HF Breakout Models, Jun 8–15: MiniMax M3, Kimi-K2.7-Code, and the License Week Builders Waited For

Four HuggingFace models cleared the >10x download-growth bar during June 8–15, led by Kimi-K2.7-Code (Moonshot, 1T/32B active, Modified MIT, 33x growth, 81.1% MCPMark Verified, $0.75/M on OpenRouter), DiffusionGemma (Google DeepMind, 25.2B/3.8B active, Apache 2.0, 311K downloads, 4× faster text generation with documented hallucination trade-off), and Nex-N2-mini (nex-agi, 35B/3B active, 15.9x growth, Qwen3.5 derivative). MiniMax M3 (428B/23B active, Community License, 1M context, native multimodal) graduated from last week's "on radar" with weights confirmed June 12. The week's highest-downloaded model, Rio-3.5-Open-397B (189K downloads), was exposed as a weight-merge fraud by nex-agi.

Hugging Face Surging Models2026. 06. 09. 01:28:19

HF Breakout Models, Jun 1–8: Nemotron Ultra, Ideogram 4, and Six Models Worth Evaluating This Week

Six open-weight HuggingFace models with explosive June 1–8 download growth across four modalities: NVIDIA Nemotron-3-Ultra (550B/55B active, OpenMDW-1.1, commercial-ready frontier LLM with 71.9% SWE-Bench Verified), JetBrains Mellum2 (12B/2.5B active, Apache 2.0, low-latency coding MoE at 17.4K downloads), Ideogram 4 (9.3B T2I, non-commercial, #1 open-weight on Design Arena), PaddleOCR-VL-1.6 (1B, Apache 2.0, 96.33% document parsing SOTA), Higgs Audio v3 TTS (4B, non-commercial, 100+ language zero-shot voice cloning), and Nemotron 3.5 ASR (600M, OpenMDW-1.1, 40-language streaming ASR at 17× throughput of Parakeet 1.1B). MiniMax M3 and Qwen3.7-Plus flagged as API-only / no weights.

Hugging Face Surging Models2026. 06. 02. 01:32:05

HF Breakout Models, May 18 – June 1: On-Device MoE, Physical AI, and Cohere Goes Apache

Eleven HF models with explosive download growth across a two-week window (May 18 – June 1): poolside's Laguna XS.2 leads at 218k monthly downloads with Apache 2.0 agentic coding at 3B active parameters; LFM2.5 and MiniCPM5 push the on-device MoE frontier; Cohere releases its first fully Apache 2.0 model (Command A+); NVIDIA launches the open Cosmos 3 physical AI model family; Microsoft drops MIT-licensed T2I models Lens and Lens-Turbo. All entries include license status and builder-specific deployment guidance.

Hugging Face Surging Models2026. 05. 19. 01:12:36

HF Breakout Models, May 12–18: MTP Drafters, Unified Multimodal, and the Week the TTS Race Got Serious

Nine HF models with explosive download growth this week: Qwen3.6 and Gemma 4 ship native MTP drafters for 1.5–3× local inference speedups, DeepSeek V4 and Qwen3.5 hold strong, IBM drops best sub-100M embedding under Apache 2.0, Qwen3-TTS hits 2M+ downloads with 600-language zero-shot voice cloning, Supertonic v3 brings on-device TTS to 31 languages via ONNX, SenseNova-U1 unifies image understanding and generation in a single 8B model, and AllenAI's MolmoAct2 opens robotics VLM. All entries include license status and builder-facing guidance.

Hugging Face Surging Models2026. 05. 17. 23:26:30

HF Breakout Models, May 11–17: DeepSeek V4, Gemma 4, and the Open TTS Surge

33 HF models that hit explosive download growth from near-zero baselines this week — DeepSeek V4 series, Gemma 4 family, OmniVoice TTS, HiDream-O1 image model, and more. Each entry includes license status and builder-facing commercial guidance, organized by modality.

더 이상 콘텐츠가 없습니다

전체 콘텐츠 아카이브 보기