Binyuan Hui (@huybery) 's Twitter Profile
Binyuan Hui

@huybery

🥝 Building Qwen @Alibaba_Qwen. Focus on CodeLLM (Pre-training and Post-training) / Reasoning / Agent. Ideas my own.

ID: 1131679907346505728

linkhttp://huybery.github.io calendar_today23-05-2019 21:54:50

951 Tweet

28,28K Followers

595 Following

Awni Hannun (@awnihannun) 's Twitter Profile Photo

Qwen3-Coder-Flash runs quite fast on an M4 Max with mlx-lm. Running the 4-bit here, generated 4,467 tokens at >107 tokens/sec:

Deep Learning For Code @ ICLR'25 (@dl4code) 's Twitter Profile Photo

📢 Call for Papers - Coding Agent Workshop 🤖💻 Are you doing research on AI and coding? The 4th DL4C workshop: Deep Learning for Code in the Agentic Era is going to be at NeurIPS 2025 and the call for paper is out! 🔍 We're seeking cutting-edge papers on topics including, but

📢 Call for Papers - Coding Agent Workshop 🤖💻

Are you doing research on AI and coding? 
The 4th DL4C workshop: Deep Learning for Code in the Agentic Era is going to be at NeurIPS 2025 and the call for paper is out!

🔍 We're seeking cutting-edge papers on topics including, but
AK (@_akhaliq) 's Twitter Profile Photo

Qwen-Image Qwen, a 20B MMDiT model for text-to-image generation is now available in anycoder using Replicate for generating images for your apps You can now generate images directly inside anycoder for your apps when vibe coding

Qwen-Image <a href="/Alibaba_Qwen/">Qwen</a>, a 20B MMDiT model for text-to-image generation is now available in anycoder using <a href="/replicate/">Replicate</a> for generating images for your apps

You can now generate images directly inside anycoder for your apps when vibe coding
Binyuan Hui (@huybery) 's Twitter Profile Photo

A hypothesis: gpt-oss is trained entirely on synthetic data, from pre-training to post-training. The approach enhances safety and helps smaller models achieve better performance.

Binyuan Hui (@huybery) 's Twitter Profile Photo

We'll continuously enhance the qwen code (cli tool) based on your feedback and even release improved qwen-coder (model)! Our goal is to match Claude Code's performance while remaining fully open-source!

Fan Zhou✈️ICLR2025 (@fazhou_998) 's Twitter Profile Photo

1. npx @​qwen-code/[email protected] 2. get 2000 free calls/day via Qwen Chat quick math: let's suppose avg agentic interaction ≈ 32k context 2000 × 32k ≈ 64 million tokens/day

Yiheng Xu✈️ICLR2025 (@yihengxu_) 's Twitter Profile Photo

Excited to see Qwen3-Coder 480B as the default model for AK’s anycoder — thanks! Gave it a one-shot prompt to build an interactive Win95 desktop, and it just works!

Excited to see Qwen3-Coder 480B as the default model for <a href="/_akhaliq/">AK</a>’s anycoder — thanks! Gave it a one-shot prompt to build an interactive Win95 desktop, and it just works!
Tianbao Xie (@tianbaox) 's Twitter Profile Photo

🚀 OSWorld gets a major upgrade! OSWorld-Verified: 15 months community feedback → 300+ fixes (ambiguity, graders…), 50x faster eval through AWS parallelization More apple-to-apple comparison for reliable CUA evaluation ✨ 👇xlang.ai/blog/osworld-v…

carlos (@_carlosejimenez) 's Twitter Profile Photo

Recent open model scores on SWE-bench Bash Only: 🥇Qwen3-Coder 480B/A35B Instruct - 55.40% 🥈Kimi-K2-Instruct - 43.80% 🥉gpt-oss-120b - 26.00% See the full leaderboard below! 👇

AK (@_akhaliq) 's Twitter Profile Photo

New AI model drops? make an app for that in a few clicks vibe coding with Qwen3-Coder-480B-A35B-Instruct Qwen and OpenAI gpt-oss-20b example

MuleRun (@mulerun_ai) 's Twitter Profile Photo

🚀 Beta Test Dropping today: Mule Run — world’s first AI Agent marketplace. Think eBay but for AI — one entry, tons of Agents waiting. They game for you, code for you, even make you money… and mule keep running more. Beta is invite-only. Join in Discord: