Binyuan Hui(@huybery)

Big congrats!! Can’t wait to see some exciting things happen!

Binyuan Hui (@huybery) — Wed, 22 Apr 2026 02:10:13 GMT

Big congrats!! Can’t wait to see some exciting things happen!

Yu Su@ysu_nlp

Introducing @NeoCognition, the agent lab for specialized intelligence.

Everyone needs experts, but human expertise does not scale.

Backed by $40M seed funding, we build self-learning agents that specialize across domains to make expertise abundant.

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬3🔄3❤️40👀11022📊8

⚡ Powered by xgo.ing

I still haven’t finished reading the Mythos system card, and now Opus 4.7 just dropped another 200+ ...

Binyuan Hui (@huybery) — Thu, 16 Apr 2026 19:19:22 GMT

I still haven’t finished reading the Mythos system card, and now Opus 4.7 just dropped another 200+ pages on my reading list. Model releases are moving faster than my reading speed...🤦‍♂️

💬3🔄2❤️55👀4241📊10

⚡ Powered by xgo.ing

bye qwen, me too.

Binyuan Hui (@huybery) — Tue, 03 Mar 2026 23:30:32 GMT

bye qwen, me too.

Junyang Lin@JustinLin610

me stepping down. bye my beloved qwen.

🔗 View Quoted Tweet

💬297🔄211❤️4691👀2642721📊771

⚡ Powered by xgo.ing

So excited to see Qwen3.5 officially out! 🎉 Congrats to everyone who made this happen. Onward and u...

Binyuan Hui (@huybery) — Mon, 16 Feb 2026 10:20:28 GMT

So excited to see Qwen3.5 officially out! 🎉
Congrats to everyone who made this happen. Onward and upward!

Qwen@Alibaba_Qwen

🚀 Qwen3.5-397B-A17B is here: The first open-weight model in the Qwen3.5 series.

🖼️Native multimodal. Trained for real-world agents.
✨Powered by hybrid linear attention + sparse MoE and large-scale RL environment scaling.
⚡8.6x–19.0x decoding throughput vs Qwen3-Max
🌍201 languages & dialects
📜Apache2.0 licensed

🔗Dive in:
Gitgithub.com/QwenLM/Qwen3.59joAT
Cchat.qwen.aiU0Rhwmodelstudio.console.alibabacloud.com/ap-southeast-1…TnHoU
Qwen Cgithub.com/QwenLM/qwen-co…nAger
Hugging Fhuggingface.co/collections/Qw…5p5um
ModelScmodelscope.cn/collections/Qw…Id57a
bqwen.ai/blog?id=qwen3.5StXaL

🔗 View Quoted Tweet

💬19🔄18❤️438👀92346📊58

⚡ Powered by xgo.ing

How to scale training environments for coding agents? Let the agent build their own! 🙌 We introduc...

Binyuan Hui (@huybery) — Wed, 04 Feb 2026 15:37:16 GMT

How to scale training environments for coding agents? Let the agent build their own! 🙌

We introduce SWE-Universe 🪐, a scalable framework that turns GitHub PRs into real-world, multilingual, verifiable SWE environments. The agent configures each environment like a human expert, with extra safeguards for reliability.

We’ve validated these environments in both mid-training and RL for Qwen3-Coder-Next, and will push environment synthesis further as a path toward agent self-improvement. Let's move on!

huggingface.co/papers/2602.02…v9F

💬14🔄46❤️364👀46651📊104

⚡ Powered by xgo.ing

I believe agents are about to enter an era of efficient execution. A promising path is to let smalle...

Binyuan Hui (@huybery) — Tue, 03 Feb 2026 16:28:15 GMT

I believe agents are about to enter an era of efficient execution. A promising path is to let smaller models interact more with the digital world and beat sheer model size through richer interaction. But it requires joint progress in data, training, and architecture. Under strict size limits, Qwen3-Coder still delivers outstanding agentic capability. Huge thanks to my team for everything!!! 💗

Qwen@Alibaba_Qwen

🚀 Introducing Qwen3-Coder-Next, an open-weight LM built for coding agents & local development.
What’s new:
🤖 Scaling agentic training: 800K verifiable tasks + executable envs
📈 Efficiency–Performance Tradeoff: achieves strong results on SWE-Bench Pro with 80B total params and 3B active
✨ Supports OpenClaw, Qwen Code, Claude Code, web dev, browser use, Cline, etc
🤗 Hugging Fahuggingface.co/collections/Qw…RJpr
🤖 ModelScmodelscope.cn/collections/Qw…zILBZ
📝 qwen.ai/blog?id=qwen3-…DYcwvd
 📄 Tech rgithub.com/QwenLM/Qwen3-C…3PWS3oi

🔗 View Quoted Tweet

💬10🔄21❤️299👀29854📊47

⚡ Powered by xgo.ing

Qwen is the fastest growing!

Binyuan Hui (@huybery) — Thu, 08 Jan 2026 02:53:32 GMT

Qwen is the fastest growing!

Nathan Lambert@natolambert

We updated the plots we use to measure the open model ecosystem at @interconnectsai, to guide The ATOM Project, and to understand what's happening.

We have ~8 plots to summarize what's happening.
First, the high level picture showing China's growing adoption lead.

🔗 View Quoted Tweet

💬5🔄4❤️76👀11182📊13

⚡ Powered by xgo.ing

Congrats to Manus! So happy for you guys! @Red_Xiao_ @peakji @hidecloud

Binyuan Hui (@huybery) — Tue, 30 Dec 2025 00:11:30 GMT

Congrats to Manus! So happy for you guys!
@Red_Xiao_@peakji@hidecloud

Manus@ManusAI

Manus is entering the next chapter: we’re joining forces with Meta to take general agents to the next level.

Full story on our blog: manus.im/blog/manus-joi…

🔗 View Quoted Tweet

💬1🔄4❤️144👀17049📊14

⚡ Powered by xgo.ing

Coding agents are moving beyond vibe coding toward real-world, enterprise software engineering. At...

Binyuan Hui (@huybery) — Sun, 14 Dec 2025 09:39:55 GMT

Coding agents are moving beyond vibe coding toward real-world, enterprise software engineering. 

Attention is shifting from frontend demos and isolated bug fixes to refactoring large codebases, improving the efficiency of complex systems, and handling long-horizon tasks. What people actually want is AI that can work continuously and reliably, delivering real productivity gains in the digital world. 

The hardest challenge is no longer writing code, but earning the trust of test engineers and addressing developers’ concerns about long-term maintenance. A new Turing test is emerging.

💬67🔄38❤️611👀62968📊168

⚡ Powered by xgo.ing

🤣 a little daydream

Binyuan Hui (@huybery) — Fri, 12 Dec 2025 12:03:50 GMT

🤣 a little daydream

TIME@TIME

2025 was the year when artificial intelligence’s full potential roared into view, and when it became clear that there will be no turning back.

For delivering the age of thinking machines, for wowing and worrying humanity, for transforming the present and transcending the possible, the Architects of AI are TIME’s 2025 Person of the Year. time.com/7339685/person…

🔗 View Quoted Tweet

💬27🔄14❤️394👀64181📊62

⚡ Powered by xgo.ing

Congrats to OpenAI on GPT-5.2! Looks like a great model.

Binyuan Hui (@huybery) — Thu, 11 Dec 2025 18:57:08 GMT

Congrats to OpenAI on GPT-5.2! Looks like a great model.

Sam Altman@sama

It is a very smart model, and we have come a long way since GPT-5.1:

🔗 View Quoted Tweet

💬49🔄30❤️1184👀136363📊154

⚡ Powered by xgo.ing

I'm thinking about how to build the next generation foundation model with Qwen.

Binyuan Hui (@huybery) — Thu, 11 Dec 2025 15:10:26 GMT

I'm thinking about how to build the next generation foundation model with Qwen.

💬37🔄13❤️416👀34850📊89

⚡ Powered by xgo.ing

Make sure you don’t miss this amazing workshop #NeurIPS2025!

Binyuan Hui (@huybery) — Sat, 06 Dec 2025 08:06:10 GMT

Make sure you don’t miss this amazing workshop #NeurIPS2025!

Deep Learning For Code @ ICML'26@DL4Code

🚀 Kicking off in TEN HOUR @ NeurIPS 2025!  Deep Learning for Code in the Agentic Era (DL4C) workshop is about to go LIVE!  Insane speaker lineup:

@gneubig@dawnsongtweets@pirroh@sidawxyz@LiangchenLuo@JustinLin610 

📍 Hall G-H | 09:00–17:00
🎁 Free customized swag for attendees!dl4c.github.ioSHzH

The future of code agents starts TODAY. See you there! 🔥

🔗 View Quoted Tweet

💬2🔄6❤️38👀9342📊7

⚡ Powered by xgo.ing

Congrats to the Qwen team!

Binyuan Hui (@huybery) — Thu, 27 Nov 2025 04:06:43 GMT

Congrats to the Qwen team!

Qwen@Alibaba_Qwen

🏆 We are incredibly honored to announce that our paper, "Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free" has received the NeurIPS 2025 Best Paper Award!

A huge congratulations to our dedicated research team for pushing the boundaries of AI.

Read more:blog.neurips.cc/2025/11/26/ann…5

🔗 View Quoted Tweet

💬2🔄4❤️219👀17221📊23

⚡ Powered by xgo.ing

😎 Qwen Chat was born from a rushed yet exciting afternoon decision. The most magical things in life...

Binyuan Hui (@huybery) — Mon, 17 Nov 2025 07:36:47 GMT

😎 Qwen Chat was born from a rushed yet exciting afternoon decision. The most magical things in life are rarely planned.

Qwen@Alibaba_Qwen

10,000,000 users creating with Qwen Chat — and we’re just getting started.
From here, let’s begin — chat.qwen.ai 🚀

🔗 View Quoted Tweet

💬26🔄15❤️273👀28986📊51

⚡ Powered by xgo.ing

So far, a big reason we still cannot do AI-trains-AI is simple. We do not have enough data about how...

Binyuan Hui (@huybery) — Sat, 15 Nov 2025 13:58:41 GMT

So far, a big reason we still cannot do AI-trains-AI is simple. We do not have enough data about how training scripts behave and how results turn out. Humans still try many runs to find the right feel. We often end up staring at result tables, debating with teammates about what really caused each change, then making new decisions. Scaling this “train→eval→thought→train” loop is very expensive. And we still lack real cold-start data from these actual workflows.

💬42🔄27❤️447👀57008📊118

⚡ Powered by xgo.ing

We’ve spent a lot of bandwidth fixing the bugs our users are running into. Please keep the feedback ...

Binyuan Hui (@huybery) — Sat, 15 Nov 2025 09:38:37 GMT

We’ve spent a lot of bandwidth fixing the bugs our users are running into. Please keep the feedback coming. We’re not in a rush to add new features. For now, we just want to make Qwen Code stable and pleasant to use.

Qwen@Alibaba_Qwen

🚀 Qwen Code v0.2.1 is here!
We shipped 8 versions（v0.1.0->v0.2.1） in just 17 days with major improvements:

What's New:
🌐 Free Web Search: Support for multiple providers. Qwen OAuth users get 2000 free searches per day!
🎯 Smarter Code Editing: New fuzzy matching pipeline reduces errors and saves tokens—fewer retries needed.
⚙️More Control: Fine-tune AI behavior with temperature, top_p, and max tokens settings.
💻 Better IDE Integration: Enhanced Zed IDE support with todo and task management tools.
📝 Cleaner Output: Tool responses now use plain text instead of complex JSON—easier for AI to understand.
🔍 Improved Search: Better file filtering (respects `.gitignore`), smarter search tools, and standardized naming.
⚡ Faster Performance: Multi-stage normalization pipeline for zero-overhead matching, better Unicode handling, and optimized output limits.
🐛 Bug Fixes: Fixed token limits for multiple models, improved cross-platform support (macOS & Windows), and better stability.

Try it now—smoother, more reliable AI codgithub.com/QwenLM/qwen-co…wj5github.com/QwenLM/qwen-co…PMyCVIskN

🔗 View Quoted Tweet

💬12🔄20❤️349👀36949📊49

⚡ Powered by xgo.ing

cool

Binyuan Hui (@huybery) — Thu, 13 Nov 2025 02:58:03 GMT

cool

Jeff Ma ✈️ ICLR’26@18jeffreyma

We’re launching SWE-fficiency to eval whether LMs can speed up real GitHub repos on real workloads! ⏱️

498 optimization tasks across 9 data-science, ML, and HPC repos — each with a real workload to speed up.

Existing agents struggle to match expert level optimizations!

🔗 View Quoted Tweet

💬2🔄6❤️46👀10130📊9

⚡ Powered by xgo.ing

Very interesting !!

Binyuan Hui (@huybery) — Wed, 05 Nov 2025 15:59:43 GMT

Very interesting !!

John Yang@jyangballin

New eval! Code duels for LMs ⚔️

Current evals test LMs on *tasks*: "fix this bug," "write a test"

But we code to achieve *goals*: maximize revenue, cut costs, win users

Meet CodeClash: LMs compete via their codebases across multi-round tournaments to achieve high-level goals

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬2🔄4❤️40👀8557📊8

⚡ Powered by xgo.ing

What’s the most exciting LLM breakthrough in the past six months?

Binyuan Hui (@huybery) — Sun, 26 Oct 2025 12:18:41 GMT

What’s the most exciting LLM breakthrough in the past six months?

💬317🔄43❤️816👀258421📊483

⚡ Powered by xgo.ing

I believe LLMs will inevitably surpass humans in coding. Let us think about how humans actually le...

Binyuan Hui (@huybery) — Wed, 27 Aug 2025 16:00:34 GMT

I believe LLMs will inevitably surpass humans in coding. 

Let us think about how humans actually learn to code. Human learning of coding has two stages. First comes memorization and imitation: learning syntax and copying good projects. Then comes trial and error: writing code, running it, fixing it, and improving through feedback. 

These two stages map closely to how LLMs learn. Pretraining (PT) compresses the entire corpus of code into parameters far beyond any human capacity. Reinforcement learning (RL)  aligns the model with execution through feedback, but at a speed and scale humans cannot match, with millions of rollouts completed in a short time. This advantage in parameter capacity and iteration means it is only a matter of time before models move ahead of humans. 

More importantly, code is not only the foundation of human productivity but also the starting point of recursive improvement. When models can write and optimize their own code, what we see is no longer just AGI but the first signs of ASI. AI4AI is on the way.

💬110🔄99❤️1014👀122094📊313