AI video now costs less than a cent a second. Vivix (@VivixLabs_HQ) just dropped W1, a real-time interactive video model, at $0.003 a second. That's around 80 W1 clips for the price of ONE Seedance 2.5 clip. The video below is their own test. Same giant cat prompt on W1, Seedance 2.5, Kling 3.0 and H3 Max. Look at the top left. That's the cheap one lol. And the price is the whole point. Real-time video keeps generating while you watch and react to it, like a story that changes based on what you do. At old prices nobody could afford to keep that running. Still early, real-time is API only for now. But give this a year and interactive video will feel as normal as scrolling YouTube.

🚀 Introducing MiniCPM-V 4.7, our new open-source MLLM series! Advancing the Performance–Efficiency Pareto Frontier. It brings strong multimodal capabilities to iOS and Android devices with extreme efficiency. Two new models: 🔹 MiniCPM-V 4.7 35B-A3B — Bring SOTA performance to local devices with newly introduced LES (Lookahead Expert Scheduling) technique, a new milestone of on-device multimodal AI. 🔹 MiniCPM-V 4.7 1B — Lightweight, efficient multimodal AI. Easier to deploy. ⚡ What's new? SOTA performance: Competitive results across 34 benchmarks spanning multimodal understanding, STEM reasoning, OCR and long-video comprehension. Fast on-device inference: LES predicts and preloads expert weights to reduce inference stalls, boosting 5× speedup on iPad Pro M4 (16GB) with byte-for-byte identical outputs. It enables MiniCPM-V 4.7 35B-A3B to achieve 17+ tokens/s on iPad Pro M4 (16GB) and 13+ tokens/s on iPhone 17 Pro Max, while costing only 3GB memory 🤯 Efficient visual processing: 4× and 16× hybrid visual token compression. Even at 16× compression, 35B-A3B scores 90.4 on OCRBench, 5.2 points ahead of Qwen3.6 35B-A3B. Easier customization: We introduce Self-Replay Fine-Tuning, a simple way to customize new target capabilities at low cost while maintaining general performance. Broad hardware support: Day-0 adaptations for Apple, Intel, AMD, Arm and Rockchip, with many more platforms supported through FlagOS! Download, deploy, and build something we haven't imagined yet. 👀👇 🔗 GitHub: github.com/OpenBMB/MiniCPM-V 🤗 Hugging Face: huggingface.co/collections/o…

I have spent more than a decade studying causality. For me, causality is about understanding the rules that govern how actions change the world, and reasoning about their consequences before acting. Scaling has made today's models remarkably capable at spotting patterns and predicting what comes next. A system that acts in the physical world needs more. It has to ask what will happen if it acts, check the outcome, and adapt when reality doesn't match the plan. In the physical world, trial and error can be costly or unsafe, and an early mistake can carry through the rest of a task. When we started Aether AI, robotics was the obvious place to begin. CRIS-0 is our first integrated system built on these ideas. It tracks the conditions that matter for a task, predicts how each action will change them, and replans when something unexpected happens. In our demo, that means taking a metal can off a plate before heating it, and telling an empty can from an unfinished drink while tidying a room. Physical AI is just our first proving ground. We're building towards causal intelligence that also works for forecasting, scientific discovery, and more. The technical write-up is linked in the comments.

Bolt is free for EVERYONE, all weekend long. And you could win $10k! Turn your idea into a product. Submit it by Oct 20 for your shot at the prize. Bolt Builders Week starts now 🏁 Details here 👉 bolt.new/builders-week

2 a.m. Furnace running, air bone-dry, and a humidifier control in the basement I've never touched. The Wi-Fi barely reaches down there. So I asked the model on my phone. MiniCPM-V 4.7: 35B total, 3B active, fully on-device in about 3 GB of memory. No cloud. One photo of the panel, one question: "Help me understand the right settings for the winter." It read the LCD 43% and told me that's already inside the 30–50% winter range. It read the dial: 1–7, "Less" on the left, "More" on the right, keep it around 4–5. Then the part I didn't expect it caught the tiny Blower Activation switch at the bottom of the panel and reminded me it has to be ON, or the humidifier does nothing. It hedged on exactly which number "Normal" is ("likely 4 or 5"), which is about what a person reading the same photo would say. How a 35B MoE model runs on a phone at all: Lookahead Expert Scheduling. Experts live on storage, and the runtime predicts which ones the next layer needs and prefetches them while the current layer computes OpenBMB reports 13+ tok/s decode in just 3 GB of memory on an iPhone 17 Pro Max, about a 5× speedup from LES. Model (open weights): huggingface.co/collections/o… Code: github.com/OpenBMB/MiniCPM-V LES write-up: github.com/tc-mb/llama.cpp-o…

Spent years watching models ace a single image and completely whiff on anything that moves or changes angle. Today we’re fixing that. Try Glimpse now, our first visual thinking model with competitive frontier performance compared to Astra on key public visual reasoning benchmarks.

There’s been so much talk about AI stealing people’s work and ending creativity. Anyone who’s made a video knows the one-shot is only a starting point. The real work happens in post. Introducing VN Studio, AI filmmaking with human authorship tracking built in, so you can copyright what you create.

Pure AI: six fingers and a vibe. Pure stock: the same photo your competitor used last week. So why choose between? Start with something real. Riff until it's right. 30M+ artist-created assets. Now editable with AI on Envato.

At Self Labs, we’ve been using agents since day one Our agentic web identity & privacy stack is already used by companies like Google Today we're launching the missing piece: Agent Vault, a secure way for agents to log in & make credit card payments Superpowers for your agents

We’re months away from AGI and the leisure economy’s about to explode Today, we’re launching Odessia, an always-on concierge for travel and experiences Prices are spiking, places are getting crowded. Odessia is your cheat code to going places Text to get started for free

Ling-3.1-flash is now open source: ~560B total parameters, ~25B active per token and up to 1M context. Built for real work, it can turn a rough brief into a page, mobile UI or interactive prototype teams can review and refine. See the demo.

Today we're launching MagicOS. And we've come to the conclusion that designers are NOT getting replaced. most of the industry moved to the same answer this year: give the agent its own canvas, let it run, hand you the output. we made a different bet: the canvas should be yours and the agent is your collaborator. that’s MagicOS. most AI design tools are a linear conversation: prompt, generate, iterate, repeat. a hundred prompts in, you might have a great final design, but the reasoning, rejected directions, and inspiration that got you there are buried in a chat log nobody will ever scroll back through. MagicOS is a persistent, spatial canvas. Your PRD, inspiration images, design system, variations, rejected directions, final picks - all of it lives together, all in one place, instead of stacked on top of itself and lost. it’s voice-first. talk to it in real time and watch work land directly on the canvas as you speak. it’s also multiplayer. your whole team works on the same canvas together. and it still runs on your design system and components because none of this matters if the output doesn’t look like your actual product. three months ago, this was internal only at Magic Patterns, my co-founder's weekend project. it’s now live for everyone. Design everything, everywhere, all at once.