Yuan3.0 Ultra ๐ฅ A 1T multimodal LLM from YuanLab
huggingface.co/YuanLabAI
โจ 64K context
โจ Enterprise-ready: RAG, summarization, Text-to-SQL
โจ 103-layer MoE w/ LAEP (49% efficiency boost)
IQuest-Coder-V1 Update! 7B & 14B series now on
@hf.co ๐ฅ
huggingface.co/collections/...
โจ 7B/14B - Base, instruct, thinking
โจ Optimized for tool use & CLI agents
โจ 128k context length
Step 3.5 Flash ๐ฅNew MoE model from StepFun
huggingface.co/stepfun-ai/S...
huggingface.co/stepfun-ai/S...
โจ Base & Base-Midtrain
โจ 196B total/11B active - Apache 2.0
โจ 256K context
โจ High-speed reasoning & agentic tasks
Qwen 3.5 Small Model Series just dropped on
@hf.co ๐ฅ
huggingface.co/collections/...
โจ 0.8B/2B/4B/9B
โจ Apache2.0
โจ 262Kโ1M token context
MiniMax M2.5 is now available on @hf.co
huggingface.co/MiniMaxAI/Mi...
โจ 229B - Modified MIT license
โจ37% faster than M2.1
โจ ~$1/hour at 100 TPS
Ovis2.6-30B-A3B๐ The latest multimodal LLM from the AIDC team at Alibaba
huggingface.co/AIDC-AI/Ovis...
โจ 64K context + 2880ร2880 resolution
โจ MoE 30B/3B active
โจ Apache 2.0
โจ โThink with Imageโ : Active visual reasoning
Ring-1T-2.5 ๐ฅ 1T reasoning model based on hybrid linear attention from Ant Group
huggingface.co/inclusionAI/...
โจ MIT license
โจ 128K -> 256K (YaRN)
โจ Hybrid MLA + Lightning Linear Attention (1:7)
โจ Agentic: Natively with Claude Code & OpenClaw
RynnBrain ๐ค a physics aware embodied brain for robots from Alibaba DAMO
huggingface.co/collections/...
โจ 2B/8B/30B (3B active)
โจ Apache 2.0
โจ Understands egocentric scenes with strong spatial awareness
โจ Tracks objects and motion over time
MiniCPM-SALA ๐ Hybrid model combining Sparse + Linear Attention from OpenBMB
huggingface.co/openbmb/Mini...
โจ 25% Sparse + 75% Linear Attention
โจ Up to 3.5ร faster inference
โจ 1M+ tokens on RTX 5090 / A6000D
โจ Apache 2.0
While Seedance 2.0โs videos are all over the timeline, DeepSeek quietly pushed a new model update in its app.
GLM-5, Ming-flash-omni from Ant Group , MiniCPM-SALA from OpenBMB, and the upcoming MiniMax M2.5 keep the heat on ๐ฅ
Spring Festival is around the corner, no oneโs sleeping!
Ming-flash-omni 2.0 ๐ New open omni-MLLM released by Ant Group
huggingface.co/inclusionAI/...
โจ MIT license
โจ MoE - 100B/6B active
โจ Zero-shot voice cloning + controllable audio
โจ Fine-grained visual knowledge grounding
New dataset ๐ฅ
UltraData-Math ๐ข math pre-training dataset released by OpenBMB
huggingface.co/datasets/ope...
โจ Apache2.0
โจ 290B+ tokens with tiered quality (L1 โ L3)
โจ Math-aware parsing that keeps formulas intact
โจ Proven to improve math reasoning (GSM8K & MATH500)
HY-1.8B-2Bit ๐ฅ A 2-bit on-device LLM released by Tencent
huggingface.co/AngelSlim/HY...
huggingface.co/AngelSlim/HY...
โจ GGUF available
โจ Near INT4 performance at just 2-bit, with full reasoning preserved
A bit late due to the flu ๐
but still very worth sharing: China open source highlights for January 2026๐ฅ
huggingface.co/collections/...
LLaDA 2.1 is out ๐ฅ MoE diffusion language models released by AntGroup
huggingface.co/inclusionAI/...
huggingface.co/inclusionAI/...
โจLLaDA2.1-mini: 16B - Apache2.0
โจLLaDA2.1-flash: 100B - Apache2.0
โจBoth delivers editable generation, RL-trained diffusion reasoning and fast inference
AI for science is moving fast๐
Intern-S1-Pro ๐ฌ a MoE multimodal scientific reasoning model from Shanghai AI Lab , now live on @hf.co
huggingface.co/internlm/Int...
โจ 1T total / 22B active
โจ Apache 2.0
โจ SoTA scientific reasoning performance
โจ Chinaโs open source AI ecosystem has entered a new phase
This final blog in the series examines how leading Chinese AI organizations are evolving ,and what this implies for the future of open source.
huggingface.co/blog/hugging...
LongCat Image-Edit-Turbo is now live on @hf.co
huggingface.co/meituan-long...
It's the distilled version of LongCat-Image-Edit from Meituan Longcat team , achieving a 10x speedup๐
GLM just entered the OCR field๐ฅ
huggingface.co/zai-org/GLM-...
โจ 0.9B
โจ MIT licensed
โจ Multimodal GLM-V architecture
โจ #1 on OmniDocBench v1.5 (94.62)
Step 3.5 Flash ๐ฅ new foundation model from StepFun AI
huggingface.co/collections/...
โจ Sparse MoE๏ผ196B/11B active
โจ Supports up to 256K context
โจ Multi-token prediction for fast decoding (100โ300 tok/s)
โจ Runs locally on consumer hardware
What a week ๐คฏ
Following DeepSeek, Kimi, Qwen, Baidu, and Ant Group,
Unitree Robotics has now released a VLA model on the
hub too!
huggingface.co/unitreerobot...
Qwen3-ASR is out๐
huggingface.co/collections/...
โจ 0.6B & 1.7B - Apache2.0
โจ 30 languages + 22 Chinese dialects, plus English accents across regions
โจ Single model for language ID + ASR (no extra pipeline stitching)
โจ Qwen3-ForcedAligner-0.6B, a strong forced aligner outperforming E2E baselines
Probably, since it's built on their own framework.
Also understandable, given that Chinese open source today is no longer just about models but entire ecosystems
Baidu just released a new VLM : PaddleOCR-VL-1.5๐ฅ
huggingface.co/PaddlePaddle...
โจ 0.9B - Apache 2.0
โจ 94.5% OmniDocBench v1.5
โจ Multilingual OCR: strong on rare characters & ancient texts
Ant Group is on fire ๐ฅ After a VLA and a depth perception foundation model, here comes a new world model!
huggingface.co/robbyant/lin...
โจ Minute-long rollouts at 16 FPS
โจ Structured camera and action control
โจ Apache2.0
LongCat-Flash-Lite๐ฅ a non-thinking MoE model released by Meituan LongCat team
huggingface.co/meituan-long...
โจ Total 68.5B / 3B active - MIT license
โจ 256k context
โจ Faster inference with N-gram embeddings
โจ LingBot-Depth:
- Metric-accurate 3D from noisy, incomplete depth
- Masked Depth Modeling (self-supervised)
- RGBโdepth alignment, works with <5% sparse depth
- Apache 2.0
Model: huggingface.co/collections/...
Paper:
huggingface.co/papers/2601....
โจ LingBot-VLA :
- Trained on 20k hours of real-world robot data
- 9 robot embodiments
- Clear no-saturation scaling laws
- Apache 2.0
Model: huggingface.co/collections/...
Paper:
huggingface.co/papers/2601....
Ant Group is going big on robotics ๐ค
They just dropped their first VLA and depth perception foundation model on @hf.co
