Adina Yakup

🧩 July 2025 - Open works from the Chinese community - a zh-ai-community Collection We’re on a journey to advance and democratize artificial intelligence through open source and open science.

🔥 July highlights from Chinese AI community

huggingface.co/collections/...

I’ve been tracking things closely, but July’s open-source wave still managed to surprise me.
Can’t wait to see what’s coming next! 🚀

Qwen/Qwen3-Coder-30B-A3B-Instruct · Hugging Face We’re on a journey to advance and democratize artificial intelligence through open source and open science.

Qwen team did it again!🔥

They just released Qwen3-Coder-30B-A3B-Instruct on the hub

huggingface.co/Qwen/Qwen3-C...

✨ Apache 2.0
✨30B total / 3.3B active (128 experts, 8 top-k)
✨ Native 256K context, extendable to 1M via Yarn
✨ Built for Agentic Coding

✨ 321B total / 32B active - Apache 2.0
✨ MFA + AFD : cutting decoding cost by up to 70% vs. DeepSeek-V3
✨ 4T image-text pretraining: strong vision–language grounding
✨ Modular, efficient, deployable: runs on just 8×48GB GPUs

Paper page - Step-3 is Large yet Affordable: Model-system Co-design for Cost-effective Decoding Join the discussion on this paper page

It’s here! After the WAIC announcement, StepFun has just dropped Step 3 🔥 their latest multimodal reasoning model on the hub.

Model: huggingface.co/stepfun-ai/s...
Paper: huggingface.co/papers/2507....

Qwen/Qwen3-30B-A3B-Thinking-2507-FP8 · Hugging Face We’re on a journey to advance and democratize artificial intelligence through open source and open science.

Qwen3-30B-A3B-Thinking-2507 🔥 latest step in scaling thinking capabilities from Alibaba_Qwen

huggingface.co/Qwen/Qwen3-3...

✨ 30B total / 3B active - Apache 2.0
✨ Native 256K context
✨ SOTA coding, alignment, agentic reasoning

Skywork-UniPic - a Skywork Collection Unified Autoregressive Modeling for Visual Understanding and Generation

Skywork UniPic 🔥a unified autoregressive multimodal model for image understanding, generation, & editing, by Skywork

huggingface.co/collections/...

✨ 1.5 B - MIT License
✨ Runs on RTX 4090
✨ Truly unified architecture

Qwen/Qwen3-30B-A3B-Instruct-2507 · Hugging Face We’re on a journey to advance and democratize artificial intelligence through open source and open science.

Qwen just released Qwen3-30B-A3B-Instruct-2507 🔥 an upgrade to the non-thinking mode model

huggingface.co/Qwen/Qwen3-3...

✨ 30B MoE / 3.3B active - Apache 2.0
✨ Strong gains in reasoning, math, coding, & multilingual tasks
✨ Native support for 256K long-context inputs

✨ 5B/14B - Apache2.0
✨ Cinematic-level aesthetics (lighting, tone, composition)
✨ Massive training data (+83% videos)→ smoother motion
✨ Supports image-only video generation, even without a prompt.

Wan2.2 🔥A video diffusion model with MoE just released by Alibaba_Wan

huggingface.co/Wan-AI/Wan2....
huggingface.co/Wan-AI/Wan2....

GLM-4.5 🔥 The largest open models yet from Zhipu AI.
Built for intelligent agents with unified capabilities: reasoning, coding, tool use.

huggingface.co/collections/...

✨ 355B total / 32B active - MIT license
✨ Hybrid modes: Thinking mode for complex tasks/ Non-thinking mode for instant replies

ScienceOne-AI/S1-Base-32B · Hugging Face We’re on a journey to advance and democratize artificial intelligence through open source and open science.

Panshi 磐石 🪨 Scientific Model by the Chinese Academy of Sciences

huggingface.co/ScienceOne-A...
huggingface.co/ScienceOne-A...

✨ 8B/32B- Apache2.0
✨ Trained on scientific data & laws across math, physics, chemistry, bio, etc.
✨ Supports 300+ tools, 170M+ papers, autonomous scientific planning

tencent/HunyuanWorld-1 · Hugging Face We’re on a journey to advance and democratize artificial intelligence through open source and open science.

TencentHunyuan released their first 3D world model: Hunyuan World 1.0 🔥

huggingface.co/tencent/Huny...

✨From a single prompt to explorable 3D scenes in minutes
✨Supports Immersive roaming / Semantic-level interactivity / Physics-ready simulation

internlm/Intern-S1 · Hugging Face We’re on a journey to advance and democratize artificial intelligence through open source and open science.

Shanghai AI Lab released their latest multimodal reasoning model: Intern-S1 🔥

huggingface.co/internlm/Int...

✨Apache 2.0
✨235B MoE LLM (Qwen3)+ 6B Vision Encoder(InternViT)
✨Pretrained on 5T tokens (50% scientific)
✨Excels at chemistry, protein folding, math, physics

GitHub - coze-dev/coze-studio: An AI agent development platform with all-in-one visual tools, simplifying agent creation, debugging, and deployment like never before. Coze your way to AI Agent creatio... An AI agent development platform with all-in-one visual tools, simplifying agent creation, debugging, and deployment like never before. Coze your way to AI Agent creation. - coze-dev/coze-studio

Coze is going open source 🔥

Coze is a full-stack platform for building & managing AI Agents built by ByteDance

✨ Coze Studio – Visual dev + plugin framework github.com/coze-dev/coz...

✨ Coze Loop – Prompt testing, evals, monitoring github.com/coze-dev/coz...

More to come. Stay tuned!

Qwen/Qwen3-235B-A22B-Thinking-2507 · Hugging Face We’re on a journey to advance and democratize artificial intelligence through open source and open science.

Qwen3-235B-A22B-Thinking-2507 🧠 new reasoning model by Qwen

huggingface.co/Qwen/Qwen3-2...

✨ 235B total / 22B active (8 experts)
✨ 256K context window
✨ Agent-ready with tool use & <think> reasoning mode

Hope the team gets some well-deserved rest this weekend after all the massive releases 🙌

inclusionAI/Ming-Lite-Omni-1.5 · Hugging Face We’re on a journey to advance and democratize artificial intelligence through open source and open science.

Ming-lite-omni v1.5 🔥 upgrade version of Ming-lite-omni, by AntGroup

huggingface.co/inclusionAI/...

✨ 20.3B / 3B active - MoE
✨ SOTA video understanding via 3D MRoPE + curriculum learning
✨ Real time speech synthesis + dialect support
✨ Enhanced multimodal generation with ID & scene consistency

✨ Highly Customizable: Supports custom terms, domain prompts, and translation memory for accurate, context-aware results.
✨ Fast and affordable: $0.5 per million tokens.

Qwen3 MT Demo - a Hugging Face Space by Qwen This tool translates text from one language to another in real-time. Users can input text and select the source and target languages from a dropdown menu. The translated text will appear in the res...

Qwen is on fire this week 🔥

Qwen3-MT🌍 Translation model supports 92 languages. Demo is available on @hf.co

huggingface.co/spaces/Qwen/...

Qwen/Qwen3-Coder-480B-A35B-Instruct · Hugging Face We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co/Qwen/Qwen3-C...

Qwen3-Coder 💻 agentic code model by Alibaba_Qwen

huggingface.co/collections/...

✨ 480B total, 35B activated MoE
✨ Agentic Coding + Browser Use → Top code model performance
✨ 256K context (up to 1M via Yarn) for repo-scale understanding

Kwaipilot/KAT-V1-40B · Hugging Face We’re on a journey to advance and democratize artificial intelligence through open source and open science.

KAT-V1 🔥 a LLM that tackles overthinking by switching between reasoning and direct answers, by Kuaishou.

huggingface.co/Kwaipilot/KA...

✨ 40B
✨ Step-SRPO: smarter reasoning control via RL
✨ MTP + Distillation: efficient training, lower cost

Anycoder - a Hugging Face Space by akhaliq Create modern, responsive web applications with HTML, CSS, and JavaScript. Users can request specific features like a todo app or a calculator, and receive complete code blocks ready for implementa...

From paper to project page in one click🚀

AnyCoder 🔥 turns research PDFs into structured, shareable project pages in seconds!

huggingface.co/spaces/akhal...

Powered by 8 SoTA open models on @hf.co

Seed-X - a ByteDance-Seed Collection A powerful open-source multilingual translation language model series, including instruction and reasoning models.

Seed-X 🔥 a suite of multilingual translation models released by ByteDance.

huggingface.co/collections/...

✨ instruction/reinforcement learning/reward model
✨ Supports 28 languages, bidirectional translation
✨ 7B with mistral architecture
✨ Excels across science, law, finance, literature & more

✨ Early-exit branches (3B–7B)
✨ Shared weights, seamless switching
✨ Fine-tuned for chat → complex reasoning

TeleAI-AI-Flow/AI-Flow-Ruyi-7B-Preview0704 · Hugging Face We’re on a journey to advance and democratize artificial intelligence through open source and open science.

TeleAI 中国电信 launched AI Flow ☁️ a new AI architecture where intelligence flows across cloud -edge - devices.

They’ve also released one of the project’s key pillars on the Hub: AI Flow Ruyi 如意 🔥 a scalable Familial Model.

Model:
huggingface.co/TeleAI-AI-Fl...
Paper:
huggingface.co/papers/2506....

Kimi-K2 - a moonshotai Collection Moonshot's MoE LLMs with 1 trillion parameters, exceptional on agentic intellegence

Kimi-K2 🔥🚀A trillion-parameter MoE model focused on long context, code, reasoning, and agentic behavior.

huggingface.co/collections/...

✨ Base & Instruct
✨ 1T total / 32B active - Modified MIT License
✨ 128K context length
✨ Muon optimizer for stable trillion-scale training

YouTube video by Pollen Robotics Your open-source companion - Reachy Mini

So excited to finally release our first robot today: Reachy Mini

A dream come true: cute and low priced, hackable yet easy to use, powered by open-source and the infinite community.

Read more and order now at huggingface.co/blog/reachy-...

POLAR - a internlm Collection We’re on a journey to advance and democratize artificial intelligence through open source and open science.

POLAR🐻‍❄️ New reward modeling by Shanghai AI Lab.

huggingface.co/collections/...

✨ 7B/8B - Apache 2.0
✨ Scalable policy discriminative pretraining
✨ Easy RLHF with minimal preference data

Skywork-Reward-V2 - a Skywork Collection Scaling preference data curation to the extreme

Skywork-Reward-V2🔥 Reward models by Skywork AI

huggingface.co/collections/...

✨ 0.6B - 8B
✨ Trained on 26M human-LLM preference pairs
✨ 0.6B > 27B in many tasks