🚨DeepSeek unveils open-source reasoning model that match Gemini-3 Pro

Microsoft's Open Agentic Model, ByteDance's Editing Model, GitHub's Design Tool, Kimi's Agentic Slides, Google's Guide.

Stay updated with today's top AI news, papers, and repos. Signup | Work With Us | Follow on X | Read on Web Hey James, Your daily briefing is ready. You can finally take a break from the AI firehose. Our algos spent the night splitting signal from noise and pulled the top news, models, papers, and repos. Here's the must-read: Summary Read time: 3 min 15 sec Top News ▸ DeepSeek releases open-source V3.2 models with strong reasoning results WorkOS ▸ Protect your AI app from advanced bots that bypass traditional detection Top News ▸ Microsoft unveils an open-weight agent that controls computers using screenshots Top Repo ▸ Vert runs file conversions locally with WebAssembly and no uploads Signals ▸ ByteDance debuts Vidi2, an AI editing model that surpasses Gemini 3 Pro on long-video tasks ▸ GitHub introduces a Figma toolkit to clarify design intent ▸ Kimi launches Agentic Slides with Nano Banana Pro and free access ▸ Google shares a Nano Banana Pro guide for consistent, high-resolution results ▸ Perplexity rolls out virtual try-on with simple photo-to-avatar workflow Top News DeepSeek introduces open-source V3.2 models with Speciale variant matching Gemini-3.0-Pro on hard reasoning 3,289 Likes DeepSeek's new V3.2 models arrive like a sequel that actually fixes the plot. The setup is simple: developers want open models that think, plan, and act with the precision of top proprietary systems. The problem is that long-context reasoning and agent workflows usually break when attention costs spike or post-training budgets run thin. The insight came from studying where open models fall short: slow attention, weak RL signals, and limited agent data. DeepSeek answers with a redesigned attention layer and a scaled reinforcement learning pipeline that treats reasoning as a first-class target. The standout moment comes from V3.2-Speciale, which reaches gold-level scores on the 2025 IMO, CMO, ICPC, and IOI and matches Gemini-3.0-Pro on complex reasoning. Key features and results: DeepSeek Sparse Attention reduces long-context compute without hurting accuracy. Reinforcement learning uses over 10% of pre-training compute to sharpen reasoning. Agent data spans 1,800 environments and 85,000 prompts for stronger generalization. V3.2-Speciale matches Gemini-3.0-Pro on demanding reasoning benchmarks. Open weights on Hugging Face support fine-tuning with LoRA or full training. TRY NOW The Battle Against Bots: How to Protect Your AI App Sponsored You see what today's automated traffic does: it executes full JavaScript, stores cookies, rotates residential IPs, and feeds CAPTCHA prompts into cheap AI solvers. Traditional detection misses more each month, and your signup flow absorbs the cost through noise, retries, and abuse. You deal with a threat profile that now looks like this: Bots that behave like full browsers Brute-force attempts masked behind rotating IPs Trial abuse that slips past basic filters Detection logic that needs constant patching WorkOS Radar addresses the entire problem with one API. It inspects real behavior, identifies automated patterns, and blocks scripted flows before they reach your core systems. It stops bots, prevents brute-force attacks, and shuts down trial abuse without extra infrastructure. The numbers hold under production-level load. Run it against your entry points and watch automated attempts drop. PROTECT YOUR APP partner with us Top Paper Microsoft introduces Fara-7B, an open-weight 7B agent that outperforms larger computer-use systems 2,159 Likes Microsoft sets the stage with a simple idea: a small model that controls a computer the way you do, by looking at the screen and deciding where to click. The problem is obvious to anyone who has tried to build web automation: real sites change, hide elements, and resist scripted bots. The insight behind Fara-7B is to skip the scripts entirely and train a model to act from screenshots like a human operator. That leads to the breakthrough: a 7B-parameter agent that reaches 73.5% on WebVoyager, competitive with larger systems that depend on multi-model orchestration. It learns from 145,000 verified trajectories and 1 million steps collected through a synthetic pipeline that explores real websites. Key Features Interprets screenshots and action history to issue precise browser actions. Uses Qwen2.5-VL-7B with 128k context for long, multi-step tasks. Learns from data generated by Magentic-One agents across diverse sites. Produces structured Playwright-style tool calls usable in automation systems. TRY NOW Top Repo Vert introduces a browser-based file converter that runs key formats on-device 3,937 Likes The Vert repo addresses a simple problem, file conversion should not require you to hand your data to strangers. Most tools upload everything first, which slows the process and moves your files off your device. Vert flips that model by running conversions for images, documents, and audio directly in your browser through WebAssembly, a technology that executes compiled code at near-native speed. The problem has been clear for years: developers need fast, private conversion without file size limits. The insight here is that modern browsers can load real conversion libraries as WebAssembly modules and run them entirely on-device. The breakthrough comes from Vert's split design: local WebAssembly for most formats and remote servers only for heavy video codecs. You can use Vert by dropping a file on the upload page or clicking to select one. Vert automatically routes the file to the correct path. TRY NOW Signals 1 ByteDance presents Vidi2, a long-video understanding and editing model beating Gemini 3 Pro and GPT-5 2,357 Likes 2 GitHub unveils the Annotation Toolkit to document design intent and eliminate preventable accessibility issues 836 Likes 3 Kimi adds agentic search and automatic file-to-slides generation across PDFs, images, and documents 1,492 Likes 4 Google presents a detailed Nano Banana Pro guide covering text rendering, character consistency, and 4K output 3,849 Likes 5 Perplexity expands its shopping tools with a virtual fitting feature using personal avatars 1,628 Likes At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We're passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are. Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We're based in the United States. By using our site, you agree to be governed by U.S. laws. Looking to promote your company, product, service, or event to 250,000+ AI developers? WORK WITH US How was today's email? Awesome Decent Not Great unsubscribe_me(): return True {"AlphaSignal": "214 Barton Springs Rd, Austin, USA"}