Marco

Agent Scaling：Building on Qwen3.5's environment scaling approach, we've aggressively expanded the quality and diversity of agentic training environments in Qwen3.7 — agentic capabilities generalize from diverse environments, just as language models do from diverse text. The figure below shows a clear and consistent improvement trajectory, with Qwen3.7-Max achieving a top-3 average ranking that approaches Claude-4.6-Opus-Max.

📣Meet Qwen3.7-Max — our latest flagship, made for the Agent Era. A versatile foundation for agents that actually get things done: 🧑‍💻 Coding agent, end to end. Frontend prototypes, multi-file refactors, real debugging — nails it. 🗂️ A reliable office and productivity assistant. Get your work done through MCP integrations and multi-agent orchestration. ⏱️ Long-horizon autonomy. 35 hours straight on a kernel optimization task — 1,000+ tool calls, zero hand-holding. 🔌 Scaffold-agnostic. Claude Code, OpenClaw, Qwen Code, or your own stack. Consistent reliability everywhere. API's up on Alibaba Model Studio. You can also take it for a spin on Qwen Studio. Go build something wild!🏃🏃‍♂️ 📖 Blog: qwen.ai/blog?id=qwen3.7 ✅ Qwen Studio: chat.qwen.ai/?models=qwen3.… ⚡️ API：modelstudio.console.alibabacloud.com/ap-southeast-1…

Introducing: Cohere Command A+ We’ve created our most powerful LLM yet, optimized it to run on as little hardware as possible, and released it open-source for all.

GOOGLE JUST SHIPPED ITS ENTIRE 2026 ROADMAP IN ONE KEYNOTE Gemini 3.5 Flash → new flagship. frontier brain, agentic, beats 3.1 pro, 4x faster Gemini 3.5 Pro → the bigger one, drops next month Gemini Omni → any input in, editable VIDEO out Gemini Spark → a personal agent that actually DOES things across your apps Daily Brief → your morning, pre-read from gmail, calendar and tasks Neural Expressive → the gemini app got a full redesign Universal Cart → one agentic cart across gemini, youtube and gmail Information Agents → search that monitors the web 24/7 FOR you Intelligent Search Box → expands as you type for real conversations Search Mini Apps → build your own dashboards inside search AI Mode → now fully running on gemini 3.5 flash Gmail Live → talk to your inbox Docs Live → write and edit docs by voice AI Inbox → gmail, organized by ai Google Keep → speak freely, it cleans it into notes Google Pics → a brand new ai image and design app Ask YouTube → search the ENTIRE youtube catalogue with answers Android XR Glasses → "intelligent eyewear," audio glasses this fall Android Halo → a live strip showing what your agent is doing Antigravity 2.0 → the agent-first dev platform, upgraded Flow + Flow Music → now standalone mobile apps

@bindureddy damn, the amount of credits it uses on chatllm is way more than 3.1

Google Makes A Come Back - Gemini Flash Early Vibes - brilliant instruction follower!! like absolutely stunning - good on agentic coding - it is NOT bench-maxxed This is genuinely a good model at a great price from Google. Overall a way better alternative to Sonnet. Will be on ChatLLM shortly

Introducing Gemini 3.5: our newest family of models combining frontier intelligence with real-world action. The first release is 3.5 Flash, our strongest model yet for agents and coding 🧵

Introducing Antigravity 2.0, a new standalone desktop application that delivers fully on that original glimpse of a truly agent-optimized experience. Rebuilt from the ground up with multi-agent teams, scheduled tasks, native voice and one-click integration with other Google products. Learn how to get started with Antigravity 2.0 👇

Welcome to Gemini 3.5 Flash, our most powerful model to date. It pushes the frontier of intelligence, speed, and cost putting 3.5 Flash in a class of its own. We spent the last 6 months making sure Flash is great for real world use cases. It's available everywhere now!

1/6 Introducing Qwen3.5-LiveTranslate: Next-gen real-time interpretation is here. 🌍 We’re breaking down language barriers with 3,500+ language pairs, ultra-low latency, visual context, real-time voice cloning, and hotword customization. Engineered to help you ship native, frictionless real-time translation experiences to a global audience.

Introducing Composer 2.5, our most powerful model yet. It's more intelligent, better at sustained work on long-running tasks, and more reliable at following complex instructions. For the next week, we’re doubling the included usage of the model.

@Alibaba_Qwen @arena qwen 4 by september at this rate

🚀🚀Qwen3.7 Preview lands on Arena ！ Here come Qwen3.7-Max-Preview & Qwen3.7-Plus-Preview. Alibaba now #6 lab in Text, #5 in Vision.⚡️⚡️ Can't wait to release Qwen3.7 series models！Stay tuned! @arena

🪄 Introducing Ardot, Tencent's AI-native design agent platform. Ardot covers the entire UI/UX workflow: → Design to code in one click. Work with CodeBuddy via MCP & Workbuddy, Cursor, Claude Code via MCP IDE. → Prompt to design. Describe it or drop an image. Get a fully editable UI draft back. → Edit anything. Select an element, describe the change, done. → Import from Figma. Full fidelity, zero migration cost. → Real-time collaboration. Comments, version diffs, team permissions built in.

Anthropic just released which jobs AI will hit hardest. Here's the full breakdown: 𝗛𝗶𝗴𝗵 𝗥𝗶𝘀𝗸 ⚠️ → Management → Business & Finance → Computer & Math → Architecture & Engineering → Life & Social Sciences → Legal → Education & Library → Arts & Media → Office & Administrative → Sales → Social Services 𝗦𝗮𝗳𝗲𝗿 𝗳𝗿𝗼𝗺 𝗔𝗜 ✅ → Installation & Repair → Construction → Agriculture → Transportation → Production → Protective Service → Food & Serving → Grounds Maintenance → Personal Care → Healthcare Support → Healthcare Practitioners The most educated, highest paid jobs on earth. All high risk. A plumber is safer than a lawyer right now. A farmer is safer than a software engineer. The world just flipped upside down. Save this. Share it with someone who needs to see it.

15 AI related accounts you should follow on Twitter: 1. @karpathy 2. @fchollet 3. @ylecun 4. @AndrewYNg 5 @rasbt 6. @dair_ai 7. @lilianweng 8. @jeremyphoward 9. @simonw 10. @_akhaliq 11. @ID_AA_Carmack 12. @gwern 13. @goodside 14 @drfeifei 15 @demishassabis Let me know who I missed guys

OpenAI spent billions on training infrastructure. Two Aussie brothers made AI training 30x faster ~ with $500K total. 🤯 Meet Daniel & Michael Han 🇦🇺 > Brothers from Sydney, Australia > Daniel was an engineer at NVIDIA > Sped up the t-SNE algorithm 2000x. Cut SVD memory in half. > Found and fixed 20+ bugs in Meta’s Llama, Google’s Gemma, Mistral, and Phi > Big AI labs missed bugs in their own models. He caught them. > Started Unsloth in December 2023 with his brother Michael > Built tools that make LLM fine-tuning 2-30x faster, with 70-90% less memory Released it 100% open source. Free for everyone. 🚀 > 64,000+ GitHub stars > 10 million model downloads every month > NASA and Canva use their code > Raised only $500K total in seed funding > Got into Y Combinator S24 > Led by two brothers with a small team of 8 shipping code While big labs burn billions, they made AI accessible to everyone. Absolute Legends 🐐

I got rejected from all the universities I applied to because my GPA was that low. I mean really really low. Fast forward to today. I'm in San Francisco surrounded by the best minds, solving real problems. Attending Y Combinator, Building my AI startup, and doing things students at schools that rejected me dream about. I'm not done yet. Not even close. My GPA hasn't come up once in a single conversation here. Nobody in SF cares about your GPA. They care about how creative you are, how fast you build, what problems you solve. So if your GPA is holding you back, you should take few steps back and change your trajectory and mindset. It doesn't matter as much as you think it does. What matters is the risks you take and what you do with them. RIP my GPA 🪦

Today we release Lighthouse Attention, a selection-based hierarchical attention for long-context pre-training that delivers a 1.4-1.7× wall-clock speedup at 98K context. It runs the same forward+backward pass ~17× faster than standard attention at 512K context on a single B200, without a custom sparse attention kernel, a straight-through estimator, or an auxiliary loss. During training, queries, keys, and values are pooled symmetrically into a multi-resolution pyramid. We then score every pyramid heads, and a top-k cascade selects a small hierarchical dense sub-sequence, and after a sorting pass that enforces causality, we use standard attention for token mixing. A brief full attention resume at the end converts the checkpoint back into a competent dense-attention model. Validated this using 530M parameter Llama-3 models across 50B tokens, with up to 1M-token benchmarks across 32 B200s under context parallelism. The work on Lighthouse Attention was led by @bloc97_, @SubhoGhosh02, and @theemozilla.

🌳Excited to introduce Orchard! 🚀 🛠️ Orchard-SWE: 67.5% on SWE-bench Verified (30B-A3B, ~3B active) 🖥️ Orchard-GUI: 68.4% avg on WebVoyager / Online-Mind2Web / DeepShop (4B!) 📬 Orchard-Claw: 73.9% pass@3 on Claw-Eval

to celebrate 3 months since lauching @lucent_ai, we're giving away 5 Codex Pro / Claude Max plans 🎁 to enter, like this post + comment which one you'd pick (codex vs claude) winners will be selected from comments in 5 days 🫶

@RaeAlisa_ @lucent_ai claude