Plus: OpenAI buys a camera startup, Qwen's 7B open-weight image model, and Runway turns video into a live stream
Sponsored by
Turn ideas into scroll-stopping videos with GoEnhance AI. Create, animate, upscale, restyle, and enhance content with AI, all in one place.
Try GoEnhance AIToday, we've got Meta turning its apps into a paid AI suite, OpenAI quietly buying a camera company, a 7-billion-parameter open-weight image model that's embarrassing closed rivals, and Runway trying to make AI video something you steer in real time.
Oh, and Americans are now using AI daily at double the rate they were six months ago. The habit is sticking.
The AI world was holding its breath for one of the largest infrastructure deals ever signed. Within 24 hours, Nvidia and OpenAI would announce a plan to deploy at least 10 gigawatts of AI compute, with Nvidia committing up to $100 billion as each gigawatt came online. A year later, we're still living inside the consequences of that handshake — in data centers, in power bills, and in every model release since.
📱 Meta One Turns Instagram, Facebook and WhatsApp Into a Paid AI Stack
Meta launched Meta One on September 15, a global subscription spanning Instagram, Facebook, WhatsApp and Meta AI with more than 50 paid features. Plans start at $2.99/month for single products and climb to $499/month for the top creator and business tier, which unlocks higher AI image and video limits, scheduling, analytics and deeper access to Meta Business Agent. Meta says testing has already produced 15 million subscriptions and trials — a serious attempt to earn money from users, not just advertisers.
📷 OpenAI Buys Glass Imaging for Over $300 Million
OpenAI acquired Glass Imaging, the computational photography startup founded by former Apple engineers who built the team behind iPhone Portrait Mode. Glass claimed its GlassAI tech could improve camera performance tenfold across phones, drones and wearables. The company had raised a $20M Series A last year at roughly a $100M valuation — this deal triples that, and it's a loud signal about where OpenAI's hardware ambitions are pointed.
🧵 Claude Code's Revamped Projects Bring Real Agent Orchestration
Anthropic rebuilt Projects in Claude Code around orchestration: Claude scopes a request, delegates work, coordinates parallel threads, reviews outputs and assembles the result — and keeps going after you close your laptop. Each discussion becomes an addressable cloud session with its own branch and repo copy. The catch: it's cloud-first, so terminal-and-local-files developers are sitting this round out.
🎥 Runway Wants AI Video to Be a Live Stream, Not a Render Queue
Instead of prompting and waiting for a finished clip, Runway is pushing toward video that streams as you steer it, built on GWM-1, its frame-by-frame world model. The company sees this stretching well beyond creative tools into robotics and autonomous driving — where a model that can imagine the next frame in real time is worth a lot more than a pretty render.
🖼️ Qwen-Image-2.1 is Alibaba's new open-weight image model that generates and edits on a single consumer GPU, with native RGBA transparency, up to ten reference images at once, and mask- or scribble-guided local edits. Weights are on Hugging Face and ModelScope — just note the research license bars commercial use without a separate Qwen license.
🔍 datasette-explain explains and validates SQL queries as you type them into Datasette, surfacing what a query actually does and where it'll be slow — handy when an LLM hands you a 40-line JOIN and a confident smile.
Coming in Hot
One API for leading AI models, at lower cost.
Open-source AI coding assistant for your IDE
AI presentation maker for business-ready PPT slides
Learn and build agentic AI systems hands-on
Infrastructure as code in any programming language
AI gaming companion that watches and reacts live
Build AI agents using natural language, no code needed
Interactive AI avatar SDK for voice agents and brands
No-code automation platform connecting 1,000+ apps
AI code review and stacked PRs for GitHub teams
Submit your AI tool to the #1 AI tools marketplace.
What it does: LLMFly AI gives developers one OpenAI-compatible API for accessing models from GPT, Claude, Gemini, Grok, and other providers. You can keep using familiar SDKs while switching the base URL and model ID, with model pricing, usage, and API costs managed through one platform.
Why It's Worth Your Time
Pricing: LLMFly AI uses pay-as-you-go pricing based on usage, with rates charged per 1 million tokens. Prices vary by model, with separate input, output, and cache rates. For example, GPT-6 Astra costs $3 input and $15 output per 1M tokens, while Gemini 3.8 Flash costs $0.30 input and $1.50 output.
Why it matters: LLMFly AI gives developers a single API layer for working with multiple leading AI models. Instead of maintaining separate integrations for every provider, teams can use one compatible interface, compare costs, switch models, and track usage from one place. This can make multi-model AI applications, coding tools, agents, and backend services easier to manage.
What it does: Mailtea is an agent-native email platform for sending transactional and marketing emails. It combines a visual email editor, REST API, SMTP, MCP server, audience management, automations, analytics, and website hosting, allowing both developers and AI agents to manage email workflows.
Why It's Worth Your Time
Pricing: Mailtea offers a Free plan with 3,000 emails/month, up to 1,000 contacts, and a 100-email daily limit. Paid plans start at $8/month for 10,000 emails, with Pro plans from $16/month for 25,000–100,000 emails and Scale from $140/month for 250,000–1 million emails. Paid plans charge $0.0007 per extra email, while the Free plan has a hard sending cap. MCP access and your own AI agent are unmetered across all plans.
Why it matters: Mailtea brings transactional email, newsletters, marketing automation, and AI-agent workflows into one platform. Its MCP integration lets AI agents handle tasks such as drafting newsletters, managing subscribers, scheduling campaigns, and checking analytics, while developers can still use familiar APIs, SMTP, SDKs, and webhooks.
🦾 Vantora Raises $100M to Mass-Produce Physical AI Startups
The studio formerly known as UP.Labs rebranded and raised $100 million to build AI companies specifically for industrial corporations — robotics, factory automation, and the unglamorous work software can't reach.
🏦 Kastle Lands $24M to Staff Banks With AI Workers
Insight Partners led a Series A for Kastle's "AI workforce" platform for consumer lending operations, with Y Combinator, Commerce Ventures and Fifth Wall joining. Back-office lending is exactly the kind of process-heavy work agents are best at — and where mistakes are expensive.
