Issue #2 ·
Cheaper DeepSeek Flash, Meta's open Muse model, Void deploys Vite apps
Issue 2 of New this week: the releases from the past week that matter if you build with AI tools, each linked to its primary source. This week: a new DeepSeek Flash model, two big open-weight releases from Meta and Black Forest Labs, an open-source way to deploy Vite apps to Cloudflare, WebSockets going GA on Vercel, and Lovable writing tests and creating Supabase projects for you.
- releaseVercel
WebSocket support in Vercel Functions is generally available
Vercel says WebSocket support in Vercel Functions is now generally available, after a public beta that started in late June. WebSocket connections follow the same limits and pricing as other Function invocations, and with Active CPU pricing, CPU is billed only for the time spent processing messages.
Why it matters: AI chat streaming and realtime multiplayer features can run on Vercel Functions without a separate realtime service.
Primary source ↗ - releaseLovable
Lovable can create a Supabase project for your app
Lovable's changelog says you can now ask Lovable to create a Supabase project for an app that has no backend: you choose the organization and region, and Lovable creates the project and connects it to your app. It works in workspaces where the built-in backend is disabled, the project's region can't be changed later, and Supabase bills the project to your own organization.
Why it matters: You can keep your app's data in your own Supabase account without wiring up the connection by hand.
Primary source ↗ - modelDeepSeek API
DeepSeek releases V4.1 Flash 1017 and opens Search and Fetch APIs
DeepSeek released DeepSeek-V4.1-Flash-1017 on its website, app and API, saying it has stronger benchmark performance, faster task completion and shorter outputs than the previous version, with weights on Hugging Face and ModelScope. DeepSeek's pricing page lists it as deepseek-flash with a 1M-token context, at $0.30 per million cache-miss input tokens and $1.20 per million output tokens at peak, half that off-peak. The release also opened Search and Fetch APIs at $0.75 per 1,000 calls each.
Why it matters: Search and Fetch calls made through Claude Code, Codex or DeepSeek Harness are free for now, but DeepSeek says billing starts on October 31 at 16:00 UTC.
Primary source ↗ - releaseLovable
Lovable now writes and runs tests for the rules you change
Lovable's changelog says it now adds a small test when you add or change a rule in your app, such as a price, a limit or a permission, and runs the tests that cover the change before reporting the work as done. The tests stay in your project, and Lovable doesn't write tests for copy, styling or markup.
Why it matters: Business rules in an AI-built app are less likely to break silently on the next prompt.
Primary source ↗ - open-sourceCloudflare Workers & Pages
Cloudflare and VoidZero open-source Void for deploying full-stack Vite apps
Cloudflare's blog announced Void, an open-source SDK, CLI and platform installed with npm install void. The SDK turns a Vite app into a full-stack Cloudflare Workers app and provisions resources such as databases, queues and AI bindings on deploy, while the self-hostable Void Platform gives a team one shared deployment environment in its own Cloudflare account, with admin-set access and resource limits.
Why it matters: Teams can let people ship vibe-coded internal apps with coding agents without handing everyone access to the cloud account.
Primary source ↗ - model
Meta releases Muse Spark Open weights
Meta released the weights for Muse Spark Open, a natively multimodal model with 816B total parameters (42B active) and a 1M-token context window, built for coding and agentic work. Meta says it runs on two Apple M5 Ultra Mac Studios with MLX, is available through API platforms including OpenRouter, Groq, Together AI and Fireworks AI, and is licensed under the Muse Model License with limited commercial terms.
Why it matters: A large open-weight coding model you can call through providers you may already use, though the license isn't fully open, so check its terms before shipping a product on it.
Primary source ↗ - open-source
Black Forest Labs releases FLUX 3 Image Klein as open weights
Black Forest Labs released FLUX 3 Image Klein, a 5.6B-parameter open-weights model for text-to-image generation, image editing and generation from up to 10 reference images. Its INT8 build runs on NVIDIA GeForce RTX GPUs with as little as 8GB of VRAM through ComfyUI or Diffusers, BFL API pricing starts at $0.009 per 1K image, and the FLUX Kommunity License allows free commercial use for companies with annual revenue under $5M.
Why it matters: Small apps can generate and edit images on a consumer GPU or a cheap API call.
Primary source ↗