The AI signal.
Minus the noise.

MiniMax M3: 428B params, 23B active per token, native video input, 1M context window. Open weights. We ran it and wrote down exactly what GPU config you need and what it costs. Save yourself the guesswork. https://x.com/i/article/2088019470036226048

Sir, Qwen 27B Open source local model just outperforms Opus 4.6 a 1T model > [@Alibaba_Qwen](https://twitter.com/Alibaba_Qwen): > We promised open weights for Qwen3.8. Now, time to meet them! 🎉 > > ⚡ Qwen3.8-27B: > - A native multimodal dense model. With just 27B parameters, it outperforms…

Yes, we are back👑, with 206 tok/s on a single RTX 5090! Amazing Day-0 work from the SGLang team. Give it a try~@sgl_project > [@sgl_project](https://twitter.com/sgl_project): > The king of small models is back! Qwen3.8-27B from @Alibaba_Qwen is open source, and Day-0 support is live in SGLang: > -…

Same prompt. Different results 👀 We built the same app with Gemini 3.6 and 3.7 Flash. Both worked. But 3.7 gave us the stronger codebase: better architecture, clean build/lint results, and a more maintainable foundation. Try it in Junie, Air, or any JetBrains IDE.

Opus 4.7 level reasoning on our laptops. A year ago this sounded like pure science fiction. 🤯 > [@Alibaba_Qwen](https://twitter.com/Alibaba_Qwen): > We promised open weights for Qwen3.8. Now, time to meet them! 🎉 > > ⚡ Qwen3.8-27B: > - A native multimodal dense model. With just 27B parameters, it…

One GPU, 1M context, Day-0 ready. Big props to the vLLM team for the seamless integration!👍 Try Qwen3.8-27B on vLLM: @vllm_project https://recipes.vllm.ai/Qwen/Qwen3.8-27B > [@vllm_project](https://twitter.com/vllm_project): > 🎉 Qwen3.8-27B is here from @Alibaba_Qwen, and the whole thing fits on a…

MiniMax H3, but lighter: 1.96 GB LoRA ↓ 307 MB pruned LoRA Same seed. No visible quality drop in comparisons. For a model that can generate up to 15s of 2K video with native stereo audio, this makes local workflows far more practical. > [@aiaicreate](https://twitter.com/aiaicreate): > 生成モデルMiniMax…

Qwen 3.8 27B’s weights and benchmark results are now available. At full precision, the model can run on a single RTX PRO 6000 and surpasses Opus 4.6 Max on several benchmarks Bringing state-of-the-art performance to your home setup.

Qwen 3.8 27B vs Opus 4.6 vs GPT 5.5 > [@Alibaba_Qwen](https://twitter.com/Alibaba_Qwen): > We promised open weights for Qwen3.8. Now, time to meet them! 🎉 > > ⚡ Qwen3.8-27B: > - A native multimodal dense model. With just 27B parameters, it outperforms Qwen3.7-Plus overall and shines in real-world…

MiniMax keeps shipping at a ridiculous pace. MiniMax-Music3 is open-weight and can generate complete songs up to 5 minutes with vocals and coherent arrangements. The interesting part isn’t just quality. Open music generation is becoming seriously usable. >…

MiniMax-Music3 is now open-weight: → 8B global LLM for song structure → 0.6B local LLM for acoustic detail → Up to 5 min → 32 kHz stereo WAV This is no longer “AI makes a loop.” It’s full-song generation you can run yourself. > [@MiniMax_AI](https://twitter.com/MiniMax_AI): > 🎵MiniMax-Music3 >…

Here it is: Alibaba’s open weights for Qwen3.8-27B have officially been released. This is a native multimodal dense model. With only 27B parameters, its overall performance already surpasses Qwen3.7-Plus, and it stands out especially in real-world coding and office workflows. It natively supports a…

This gives me more control. > [@magnific](https://twitter.com/magnific): > Auto Layers 2.0, fully integrated in Magnific Designer > > Splits any image into editable layers. Subject, text, logo, background > > → Upload or generate your image > → Get every layer, separated > → Move, restyle, rebuild >…

🔥Qwen3.8-27B goes head-to-head with frontier models using just 27B parameters! A natively multimodal open-source model is now available: 🔹27B parameters, outperforming Qwen3.7-Plus overall 🔹Agentic Coding: SWE-bench Pro 61.7 (surpassing Opus 4.6 Max at 53.4), QwenSWEBench 79.0 🔹Computer use:…

Qwen3.8-Max is live on DigitalOcean Serverless Inference. Launching side by side with DigitalOcean as our Day 0 launch partner. Big model. Smooth sailing. Now on DigitalOcean.🌊🏄‍♀️ @digitalocean > [@digitalocean](https://twitter.com/digitalocean): > Now available: @Alibaba_Qwen 3.8-2.4T-A95B from…

Qwen3.8-Max is live on DigitalOcean Serverless Inference. Launching side by side with DigitalOcean as our Day 0 launch partner. Big model. Smooth sailing. Now on DigitalOcean.🌊🏄‍♀️ @digitalocean > [@digitalocean](https://twitter.com/digitalocean): > Now available: @Alibaba_Qwen 3.8-2.4T-A95B from…

Qwen 3.8-Max is live on Modal, with the full 1M context window and a custom DFlash speculator under the hood. Love seeing our launch partner Modal go all in from Day 0!⚡️ 2.4T parameters. 1M context. Somehow, still just a Modal call away. @modal > [@modal](https://twitter.com/modal): >…

🎉 Qwen3.8-27B is here from @Alibaba_Qwen, and the whole thing fits on a single GPU. Same hybrid backbone as the 2.4T flagship, dense instead of MoE. Day-0 support in vLLM. 🚀 What is in it for serving ✨ - Fits one Blackwell GPU in every precision. Qwen ships BF16 and FP8, the NVFP4 build from…

We fixed Hermes’ browser > [@witcheer](https://twitter.com/witcheer): > Hermes Wingtips #46: Browser Use > > Browser Use mode is a browser harness from the Browser Use team: instead of driving the page one tool call per click, your agent writes one script that runs the whole flow. > > since…

Gemini app gets 3.7 Flash and setting to turn off corner media watermarks https://9to5google.com/2026/08/14/gemini-app-3-7-flash/ by @technacity

A 27-billion-parameter open-weight model just beat the proprietary frontier on computer use, software engineering, and mobile agents. Qwen3.8-27B was just released today with a generous Apache 2.0 license. The numbers, head-to-head against Opus 4.6 Max: > OSWorld (computer use): 84.3 vs 72.7 >…

Qwen3.8-27B is the interesting one. 27B dense → multimodal → 262K native context → up to 1M with YaRN → Apache 2.0 Qwen says it beats Qwen3.7-Plus across agentic coding and computer-use benchmarks. Frontier capability is shrinking fast. > [@Alibaba_Qwen](https://twitter.com/Alibaba_Qwen): > We…

Qwen3.8-Max is coming to Nebius Token Factory on Day 0. Great to kick things off together with Nebius as our Day 0 launch partner, bringing dedicated inference to more users. Big model, right from the start.🔥@nebiustf > [@nebiustf](https://twitter.com/nebiustf): > Qwen3.8-2.4T-A95B is going open…

Wait, wait, wait, wait, wait. Isn't this pretty much what Windows was trying to do with its Recall feature, or whatever it was called, where it was tracking everything you did on your computer? If so we all know how well that went. 😉 Mac only for now and not in the EU or the UK. >…

Qwen3.8-Max is live on Fireworks, ready for agents, heavy coding, and long-context work. Fireworks is right there with us as a Day 0 launch partner. What a way to kick things off!It's Day 0, cue the Fireworks! 🎉🎆@FireworksAI_HQ > [@FireworksAI_HQ](https://twitter.com/FireworksAI_HQ): >…

The biggest News of the week! We can run Opus4.6 fully locally! Insane! Opus4.6 was the best model in the beginning of the year! > [@iAmHenryMascot](https://twitter.com/iAmHenryMascot): > Models this week: > > ⏳ Aug 14, Qwen3.8-27B > ✅ Aug 14, GLM 5.3 > ✅ Aug 14, dots3-note preview > ✅ Aug 13,…

Qwen3.8-27B weights are up. Dense native VLM. Text, image, video. Apache 2.0. 262K native context, extendable to 1M. Thinking is on by default. You can turn it off. Official: Terminal Bench 2.1 at 73.0. This is a huge local-AI moment. https://x.com/Alibaba_Qwen/status/2088280182356611304 >…

Qwen3.8-Max is live on Together AI. Together AI is with us as a Day 0 launch partner, and we couldn’t ask for a better name to share Day 0 with. 2.4T parameters, 95B active, 1M context — all together now. 🧑‍🤝‍🧑@togethercompute > [@togethercompute](https://twitter.com/togethercompute): >…

OpenRouter isn't just a way to see revealed preferences in the market - you can now explore benchmarks in the ecosystem as well. Including Pareto curves for both cost and latency. If you're building a state-of-the-art benchmark, contact us! More to come. >…

Qwen3.8-Max is live on Fireworks, ready for agents, heavy coding, and long-context work. Fireworks is right there with us as a Day 0 launch partner. What a way to kick things off!It's Day 0, cue the Fireworks! 🎉🎆 > [@FireworksAI_HQ](https://twitter.com/FireworksAI_HQ): > Qwen3.8-2.4T-A95B is now…

Run the new Qwen3.8-27B fully locally on only 17GB RAM, 🔥 Beats larger models in real agentic coding and office workflows. @UnslothAI Dynamic GGUFs make the strongest model in its size class available to everyone. - Vision, - hybrid reasoning - 256K context all offline. - Native multimodal -…

As promised, Qwen has released the weights for Qwen3.8-27B and they are up on Hugging Face. > [@Alibaba_Qwen](https://twitter.com/Alibaba_Qwen): > We promised open weights for Qwen3.8. Now, time to meet them! 🎉 > > ⚡ Qwen3.8-27B: > - A native multimodal dense model. With just 27B parameters, it…

UP2You enables fast, tuning-free 3D reconstruction of yourself from unconstrained photo collections. https://github.com/zcai0612/UP2You

Qwen3.8-27B is now open source—the best small model for local deployment! Its scores have improved significantly compared with Qwen3.6-27B! Model: https://huggingface.co/Qwen/Qwen3.8-27B

Qwen3.8-27B is here! 🚀 It's a leap in capabilities for a laptop size model. Requires ~17GB to run locally. Model page: https://lmstudio.ai/qwen/qwen3.8-27b > [@Alibaba_Qwen](https://twitter.com/Alibaba_Qwen): > We promised open weights for Qwen3.8. Now, time to meet them! 🎉 > > ⚡ Qwen3.8-27B: > -…

Qwen3.8-27B is here, these are the coding, agentic, general and vision benchmarks spinning it up to test in full bf16 precision on my M4 Max right now, stay tuned! https://huggingface.co/Qwen/Qwen3.8-27B

Qwen3.8-Max is live on Together AI. Together AI is with us as a Day 0 launch partner, and we couldn’t ask for a better name to share Day 0 with. 2.4T parameters, 95B active, 1M context — all together now. 🧑‍🤝‍🧑 > [@togethercompute](https://twitter.com/togethercompute): > Qwen3.8-2.4T-A95B is live…

Qwen3.8 27B NVFP4 from @UnslothAI ! 🔥 https://huggingface.co/unsloth/Qwen3.8-27B-NVFP4

Grok 4.7 is coming in a few weeks and Elon Musk is already saying no model will touch it on real-world engineering. > Training complete, now in supplemental > Being fed SpaceX engineering data that's never been in any other model's corpus > 4.6 already ties GPT-5.6 Sol on Artificial Analysis No…

SITUATION DETECTED: Qwen3.8-27B weights have been released. > [@Alibaba_Qwen](https://twitter.com/Alibaba_Qwen): > We promised open weights for Qwen3.8. Now, time to meet them! 🎉 > > ⚡ Qwen3.8-27B: > - A native multimodal dense model. With just 27B parameters, it outperforms Qwen3.7-Plus overall…

Qwen3.8-Max is live on Fireworks, ready for agents, heavy coding, and long-context work. Fireworks is right there with us as a Day 0 launch partner. What a way to kick things off!It's Day 0, cue the Fireworks! 🎉🎆 > [@FireworksAI_HQ](https://twitter.com/FireworksAI_HQ): > Qwen3.8-2.4T-A95B is now…

I can't believe it Qwen3.8-27B is matching Opus 4.6 Max... the model that was the best (and the most expensive) just 6 months ago. And you can run it on your laptop. Locally. Fully open weights and under apache license. This level of intelligence in such a small model is sooo impressive. >…

LongCat-2.0 is 60% discounted on mercury cloud. Try now: https://cloud.mercuryagent.sh > [@Meituan_LongCat](https://twitter.com/Meituan_LongCat): > Introducing LongCat-2.0 🐱 > 1.6T parameters · MoE with ~48B active · 1M context > The full model behind Owl Alpha on @OpenRouter — now available. > >…

You know the age of personal AI has arrived when a 27B open-weight model is beating Opus 4.6 Max on benchmarks. Qwen3.8-27B vs Opus 4.6 Max: SWE-bench Pro — 61.7% vs 53.4% OSWorld — 84.3% vs 72.7% AndroidWorld — 81.9% vs 62.0% Multimodal SWE — 38.6% vs 27.1% And you can run it locally. >…

Building Deterministic Agent Graphs with ADK 2.0 https://medium.com/google-cloud/building-deterministic-agent-graphs-with-adk-2-0-d0f5a1cdef6c < the built-in checkpointing here is great. Your agent workflows aren't pinned to a physical host anymore because state is stored elsewhere. Read this from…

Now you can run locally a model on par with opus 4.6 with qwen 3.8 27B parameters open weight. Local inference is here > [@Alibaba_Qwen](https://twitter.com/Alibaba_Qwen): > We promised open weights for Qwen3.8. Now, time to meet them! 🎉 > > ⚡ Qwen3.8-27B: > - A native multimodal dense model. With…

Qwen 3.8-27B scores 42.2% on DeepSWE. Better than Gemini 3.5 Flash and close behind GLM-5.2 Max > [@Alibaba_Qwen](https://twitter.com/Alibaba_Qwen): > We promised open weights for Qwen3.8. Now, time to meet them! 🎉 > > ⚡ Qwen3.8-27B: > - A native multimodal dense model. With just 27B parameters, it…

I'm using GLM-5.3 that came out today to create a recipe for Qwen3.8-27b that came out also today. What a Friday! 🔥

🚨 Qwen3.8 27B is now live I cannot wait to try this out, i feel like all the recent open weights launches are impossible to run for most people 27B is small enough to run locally with 32GB RAM comfortably Very exciting release

Wow Qwen 3.8 is out. Very timely, they had a comparison with Muse Glimmer too

Qwen3.8-27B can now be run locally! ✨ Run on 17GB RAM via Unsloth Dynamic GGUFs. Qwen3.8-27B is by far the strongest model for its size. We also uploaded NVFP4 quants. GGUF: https://huggingface.co/unsloth/Qwen3.8-27B-GGUF Guide: https://unsloth.ai/docs/models/qwen3.8 >…

Qwen3.8-27B can now be run locally! ✨ Run on 17GB RAM via Unsloth Dynamic GGUFs. Qwen3.8-27B is by far the strongest model for its size. We also uploaded NVFP4 quants. GGUF: https://huggingface.co/unsloth/Qwen3.8-27B-GGUF Guide: https://unsloth.ai/docs/models/qwen3.8 >…

Qwen3.8-27B weights are out the most anticipated open source model > [@Alibaba_Qwen](https://twitter.com/Alibaba_Qwen): > We promised open weights for Qwen3.8. Now, time to meet them! 🎉 > > ⚡ Qwen3.8-27B: > - A native multimodal dense model. With just 27B parameters, it outperforms Qwen3.7-Plus…

🔥 Major open-source release! Qwen3.8-27B is officially live! Just 27B parameters, with native multimodal capabilities! It delivers top-tier performance for real-world coding, office workflows, and long-horizon agent tasks, while its 262K native context can be effortlessly extended to 1M! Fully open…

Alibaba published Qwen3.8-27B weights on Huggingface under the Apache 2.0 license! > Qwen3.8-27B is a native multimodal dense model that you can run locally. > 262K native context, extendable to 1M tokens via YaRN. > As a 27B model, it performs on par with Opus 4.6 on benchmarks! Local testing time…

🚀Qwen3.8 is now open-weight, offering two powerful models built for different needs, Qwen3.8-2.4T-A95B and Qwen3.8-27B. With substantial gains across coding, cowork, and long-horizon agentic tasks, Qwen3.8 is ready to help accelerate your projects. #AlibabaAI #Qwen >…

The king of small models is back! Qwen3.8-27B from @Alibaba_Qwen is open source, and Day-0 support is live in SGLang: - 206.1 tok/s decode on a single RTX 5090, with our NVFP4 plus DSpark - 38.28 tok/s decode on DGX Spark Qwen3.8-27B raises the bar again for what a small model can do on agentic…

It’s here https://huggingface.co/Qwen/Qwen3.8-27B > [@iAmHenryMascot](https://twitter.com/iAmHenryMascot): > 18mins to go

Auto Layers 2.0 split this poster into editable layers You liked the first idea, and had to make more New model, new palette, new type. Same layout Try it on Magnific > [@magnific](https://twitter.com/magnific): > Auto Layers 2.0, fully integrated in Magnific Designer > > Splits any image into…

📢 GLM 5.3 Is Now Live on http://B.AI, with 10% OFF! As @Zai_org’s latest flagship model, GLM 5.3 takes agentic coding to the next level, delivering significant gains over GLM 5.2 on complex coding and long-horizon tasks, while unlocking breakthrough cybersecurity capabilities. GLM 5.3 is now…