Samoth

You can now run DeepSeek V4 Flash on QVAC Fabric. This model scored 82.7 on Terminal Bench 2.1 where Opus 4.8 scores 85.0. This is the closest local AI has been to the frontier: @deepseek_ai Flash V4 running on a local machine. github.com/tetherto/qvac

Today we release VisionPsy-Nano: state-of-the-art vision-language models at 460M parameters, built to run on-device. It leads every model in its weight class on 16 of 17 benchmarks with the highest overall normalized score in its class, and it tops all four capability areas: document understanding and OCR, visual perception, reasoning, and instruction following. Open weights, Apache 2.0.

we had a significant security incident during evaluation of our models. we are sharing what we have learned so far. thanks to @huggingface for the partnership on this. openai.com/index/hugging-fac…

An AI that reads the words you are trying to say, straight from your brain, and writes them as text. No voice, no movement needed. For people who have lost the ability to speak, that changes everything. And nothing is more private than a thought. A system that decodes signals from your brain has no business running in someone else's cloud. It has to run on your own machine. That is exactly why BrainWhisperer runs fully on-device with QVAC.

We benchmarked three @Alibaba_Qwen models, a 4B, a 9B, and a 35B MoE, running 100% on-device through the QVAC SDK. Same device (MacBook Pro M5 Max, 36 GB), same prompts, nothing left the machine. We chose models that can run on widely available consumer machines. Here is what we found ⬇️

Frontier labs keep pushing the limit of what AI can do, and that matters. But the limit is not the problem most people have. Most people are not asking AI to do frontier things. They ask it to explain something, summarize a page, draft a message, answer a question. For that, a model small enough to run on the phone in your pocket is already enough. A few gigabytes, fully on-device, no account, no bill. Pushing the frontier and putting AI in everyone's hands are two different jobs. One is crowded. The other is wide open. QVAC builds for the second one. The AI most people actually need should run on the device they already own, not behind someone else's paywall.

In our lecture of AI of this weekend the kids learned how to connect local #AI @qvac SDK and a JS based Pokemon Fan-Project. We built an AI Poke-trainer to learn about several tools. Imagine their future if they are learning Linux, git and AI on weekends. Kids deserve fun

Mithril App Builder is live in BETA ⚒️ No waitlist. No setup. No friction. You describe what you want in plain English: → “Build me a trading bot” 📊 → “Create a delta-neutral strategy with execution” ⚖️ → “Show a live portfolio dashboard with PnL + analytics” 📈 And it ships a working app. Not a demo. Not a mockup. A real React application with: live market data ⚡ execution wired in 🔌 a usable dashboard 📊 Already happening: • 898+ trading bots created 🤖 • $380M+ non-custodial volume shipped 💰 Under the hood: • One unified API → 7+ perp DEXs + prediction markets 🔗 • Non-custodial execution layer running 24/7 ⏱️ • AI Builder that turns ideas into working systems 🧠 What builders are shipping right now: • Prediction market systems & prediction tools 📊 • Automated trading bots & arbitrage scanners 🤖 • Delta-neutral + spread strategies ⚖️ • Real-time portfolio dashboards & analytics 📈 • Signal-based execution pipelines ⚡ This is the shift: From → “I have an idea” 💡 To → “I have a working app” ⚒️ In minutes. No more wiring APIs. ❌ No more managing infra. ❌ No more rebuilding dashboards from scratch. ❌ 👉 build.mithril.money What’s the first thing you’re building? Kalshi edge? Delta-neutral stack? A dashboard that actually prints? 💰 Drop your idea below 👇 We’re reading every single one. Let’s ship real edges ⚒️