We advance the development of ASI and foster open source collaboration towards a smarter future. Discord: discord.gg/mnPyh8ZUEc
- Tweets 1,127
- Following 26
- Followers 24,964
- Likes 265
Pinned Tweet
Introducing Qwen3.7-Plus, the latest flagship addition to the Qwen3.7 model series.
Built to bridge the gap between visual perception and terminal execution, it serves as a versatile foundation to power your diverse multimodal agent workflows. 🚀
Key highlights:
• Multimodal Interactive Hybrid Agent: Enables unified GUI & CLI operation across visual and text tasks.
• Versatile Coding Agent & Productivity Assistant: Handles full-modality input to supercharge your daily productivity.
• Visual Agent: Deepens agent intelligence with advanced perception, reasoning, grounding, and search-augmented QA.
• Cross-Harness Generalization: Delivers consistent, robust performance across diverse agent frameworks.
The next generation of Qwen3.7 family has arrived to support your AI agent workflows.
52
39
615
8,541,216
Hello, creators and builders,
This week brings us new research on language world models for agents, and a new T2I model benchmark that closes the gap between scores and real-world performance. Behind them, our community continues to deliver practical tools for you.
Let’s dive in.
open.substack.com/pub/tongyi…
3
3
53
3,836
A new approach to general agent capabilities: Qwen-AgentWorld. 🤖
While LLMs are typically trained to act in environments, Qwen-AgentWorld is trained to model the environments themselves across 7 domains (MCP, Search, Terminal, SWE, Web, OS, Android).
The research shows that using Controllable Sim RL and LWM warm-up significantly improves agentic tasks without specific fine-tuning, achieving top performance on AgentWorldBench.
Check out the full details in the thread below! 👇
📣📣 Meet Qwen-AgentWorld — a native language world model that simulates 7 agent environments (MCP, Search, Terminal, SWE, Web, OS, Android) within a single model. Environment modeling is the training objective from day one, not a post-hoc adaptation.
🤔 LLMs are trained to be better agents — better at acting in environments. But nobody has trained them to model the environments themselves.
🗺️ Our roadmap: investigate how language world modeling can push the boundaries of general agent capabilities, along two routes:
1️⃣ Build a foundation model for environment simulation — outperforming Claude Opus 4.8 and GPT-5.4 on AgentWorldBench
2️⃣ Investigate how world modeling enhances agent training:
🔬 Controllable Sim RL (agentic RL with LWM as environments) surpasses training in real environments
🧠 Learning to predict environments (LWM warm-up) makes agents stronger — remarkably, even without any agent-specific training, this predictive knowledge transfers to agentic tasks with zero fine-tuning
📑 Paper: arxiv.org/abs/2606.24597
📖 Blog: qwen.ai/blog?id=qwen-agentwo…
💻 GitHub: github.com/QwenLM/Qwen-Agent…
🤗 HuggingFace: huggingface.co/collections/Q…
🧩 ModelScope: modelscope.cn/collections/Qw…
9
24
206
16,483
AI excels in the virtual world, but acting in the physical world is a whole different challenge. Why is it so hard to get a robot to fetch an egg without freezing? 🤔
Check out our new explainer video on Embodied Intelligence. We discuss the gap between robot "Thinking" and "Acting,"
18
23
279
2,842,170
What if science had a common language?
Proteins, molecules, materials, and reactions all seem different. What if they could be represented through a shared scientific language?
Join us for the first live episode of Ready to Share, where the author of LOGOS discuss the ideas behind the research.
📅 June 30, 13:00 UTC
[LIVE] Ready to Share #01
LOGOS: Teaching AI a Common Language for Science
Fresh research, explained by the people who built it.
nitter.app/i/broadcasts/1vJpPPWlE…
3
3
14
2,224
Hello, creators and builders,
This week marks significant strides across our entire ecosystem—from AI for science to fundamental infrastructure, real-world robotics and community trending community-tuned models.
Let’s dive in.
open.substack.com/pub/tongyi…
1
2
34
3,543
Huge congratulations to @ZvecAI for hitting #1 Trending on GitHub in C++! 🚀
This milestone comes right after the release of Zvec 0.5.0, which introduced significant updates including native full-text search, the new DiskANN on-disk index, a brand-new Ecosystem with Go/Rust SDKs, and RISC-V support.
Check out the project and join the community to see why it's trending!👇
Github: github.com/alibaba/zvec
1
6
54
3,408
We're excited to share our latest research: LOGOS.
AI for Science has produced powerful models for proteins, molecules, materials, reactions, and other scientific objects.
Every new task often requires a new model, a new training pipeline, and a new set of assumptions. Knowledge stays siloed.
This raises a question:
Can scientific data be modeled like language, through a shared generative framework?
To investigate this, we introduce LOGOS, a general-purpose generative foundation model for the natural sciences.
We'll also host an upcoming live session with the authors. More details soon.
A short thread 🧵
37
58
431
5,781,051
Key innovations:
🔹 Unified scientific grammar: LOGOS maps scientific objects and their relationships into a shared discrete token space. Instead of explicitly modelling atomic-level coordinates, it uses sequential representations to encode spatial contact and constraint information.
🔹 Pretraining–generation alignment: LOGOS aligns next-token prediction during pretraining with generation objectives in downstream tasks, reducing the gap between learning and application.
🔹 Cross-domain knowledge transfer: The results prove that the unified grammar is not merely a representational unification, but rather achieves genuine knowledge interoperability and synergy across modalities by constructing deep semantic bridges in a shared grammar space.
1
3
14
2,080
Future AI for Science systems may benefit from aligning scientific foundation models with LLMs through shared architectures, shared training paradigms, and shared inference infrastructure, rather than evolving as entirely separate stacks.
To support future research, we are open-sourcing the model, code, and technical paper👇
🤗 HuggingFace: huggingface.co/LOGOS-Hub
💻 GitHub: github.com/LOGOS-Hub/LOGOS
📄 Paper: arxiv.org/abs/2606.16905
4
26
4,349
We are pleased to highlight an excellent community model from developer : Qwen3.6-27B-MTP-pi-reasoning-GGUF.
Built on our Qwen3.6-27B base model, this release focuses on optimizing automated programming and debugging workflows for local coding agents.
If you are exploring local AI coding assistants, we encourage you to test this model in your environment.
45
139
1,235
114,430
A look at the new Qwen-Robot Suite-a full stack for embodied intelligence.
They serve as a low-level toolkit for general-purpose agentic systems to act in the physical world.
📣 Introducing the Qwen-Robot Suite — Qwen-RobotNav, Qwen-RobotManip, Qwen-RobotWorld, three foundation models, a full stack for embodied intelligence.
🧭 Qwen-RobotNav — the gateway to mobility.
• Unifies 5 navigation tasks in one model: instruction following, point-goal, object-goal, target tracking, autonomous driving
• Controllable observation protocol
• Tool interface for agentic systems
🤖 Qwen-RobotManip — the foundation of interaction.
• Unified state-action space across heterogeneous robots
• Camera-frame delta poses for coherent cross-embodiment training
• Pretrained on a 38,100+ hour open-source corpus
🌍 Qwen-RobotWorld — infinite worlds for physical agents.
• Single world model, 20+ embodiments
• Natural-language action interface
• Predicts physically grounded futures across manipulation, driving, and navigation
Each model is independently useful, and could be composed as physical-world tools.Together, they form the low-level toolkit for general-purpose agentic systems that don't just see the world, but act in it.
📷 Blog:
qwen.ai/blog?id=qwen-robotsu…
📖 Report：
Qwen-RobotNav: qianwen-res.oss-accelerate.a…
Qwen-RobotManip: qianwen-res.oss-accelerate.a…
Qwen-RobotWorld： qianwen-res.oss-accelerate.a…
1
6
40
3,441