# LU Labs > LU Labs builds an all-in-one local AI studio: chat, coding agent, image and video generation in one free desktop app for Windows, macOS and Linux. Everything runs on your own hardware, with no account needed. The Mac build is a beta for Apple Silicon, macOS 14 or newer; there is no Intel build. LU Cloud (Hosted 19€/mo, Hosted Pro 49€/mo, Hosted Max 99€/mo, or one-time credit packs) runs the same open-weight models on managed GPUs, for phones, light laptops and models too big for home VRAM. ## Current cloud facts Updated 2026-09-14. Hosted includes 11 Flash models without credits; Pro and Max include 12. Unlimited prompt counts, up to 500,000 combined input and output tokens per account per UTC day in app sessions. All 48 chat models remain available on every plan. GLM 5.3 Flash uses credits on Hosted and is included in the daily allowance on Pro and Max. API keys and accounts without an active subscription always use credits. Image, video, API and metered text share one monthly wallet. Per-model output estimates are alternatives, not additional allowances. - [Product facts](https://lu-labs.ai/product-facts): Current prices, per-plan allowances and model-specific output estimates. - [Product facts JSON](https://lu-labs.ai/product-facts.json): The same facts in machine-readable form. ## Pages - [Current product facts](https://lu-labs.ai/product-facts): Subscription prices, included Flash models by plan, API billing and per-model image and video budgets. - [Home](https://lu-labs.ai/): Overview of the studio (chat, code, image, video), the privacy guarantees and pricing at a glance. - [Features](https://lu-labs.ai/features): The four studios in detail: private chat, coding agent, image generation and video generation. - [Cloud](https://lu-labs.ai/cloud): LU Cloud: open-weight chat, image and video models on managed NVIDIA H100, A100 and B200 class GPUs. - [Pricing](https://lu-labs.ai/pricing): Self-Host free forever, hosted plans, and one-time credit packs. Every plan reaches the same model catalog. - [How it works](https://lu-labs.ai/how-it-works): Detects the local runtimes you already have (Ollama, LM Studio, vLLM, llama.cpp, ComfyUI), loads a model into your memory, generates on-device. - [LU Cloud handbook](https://lu-labs.ai/guide): Using the hosted studio in a browser: sign up, chat, Create and Compare, plans, credit packs, the Flash allowance, API keys and account settings. - [API reference](https://lu-labs.ai/docs/api): The LU Cloud API: OpenAI compatible chat and images, every Create job over one endpoint, quotes, uploads, the content setting per key, limits and errors. - [Download](https://lu-labs.ai/download): The free desktop app for Windows, macOS and Linux. The Mac build is a beta for Apple Silicon, macOS 14 or newer. - [FAQ](https://lu-labs.ai/faq): Privacy, quotas, billing and the difference between local and hosted. - [Blog](https://lu-labs.ai/blog): Guides and engineering notes on open-weight models, local AI and the LU Cloud catalog. - [DeepSeek V4.1 Flash online](https://lu-labs.ai/deepseek-v4-1-flash-online): Running DeepSeek V4.1 Flash in a browser or from an OpenAI-compatible client, on the day the open weights landed and long before any runtime can load them. - [Kimi K3.1 Pro online](https://lu-labs.ai/kimi-k3-1-pro-online): Running Moonshot Kimi K3.1 Pro, the post-trained successor to Kimi K3, in a browser or from an OpenAI-compatible client, before its open weights are out. - [Kimi K3 online](https://lu-labs.ai/kimi-k3-online): Running Moonshot Kimi K3 in a browser or from an OpenAI-compatible client, without downloading 2.8 trillion parameters. - [Run Qwen 3.8 without a GPU](https://lu-labs.ai/run-qwen-3-8-without-gpu): Both Qwen 3.8 flagships on managed GPUs, answering in the browser. - [FLUX 2 online without a GPU](https://lu-labs.ai/flux-2-online-without-gpu): Flux 2 image generation with no 24 GB card and nothing to install. - [LM Studio alternative for Mac](https://lu-labs.ai/lm-studio-alternative-for-mac): The models a Mac cannot hold, in the browser, next to the desktop app for Apple Silicon. - [Impressum](https://lu-labs.ai/impressum): Provider identification under § 5 DDG and § 18 MStV. - [Delete your LU account](https://lu-labs.ai/delete-account): How to delete an LU account in the app or on the web, what is deleted and what is kept. - [LU support](https://lu-labs.ai/support): How to reach us about the LU apps and lu-labs.ai, with the account and privacy pages. - [Privacy](https://lu-labs.ai/privacy): Privacy notice: no telemetry and no prompt logging in local mode. - [Terms](https://lu-labs.ai/terms): Terms of service. ## Blog - [DeepSeek V4.1 Flash 1017 replaces V4.1 Flash in LU Labs Cloud](https://lu-labs.ai/blog/deepseek-v4-1-flash-1017-in-lu-labs-cloud): DeepSeek's October refresh of V4.1 Flash is live in LU Labs Cloud. Old picks and API calls move over on their own, the catalog price is lower, and here is what DeepSeek says changed. - [Qwen-Image 2.1 Runs Locally: One Open Model That Draws and Edits](https://lu-labs.ai/blog/run-qwen-image-2-1-locally): Qwen-Image 2.1 is one model for text to image and for editing. What it needs in VRAM, what its research licence forbids, and how it runs in LU Labs 3.0.1. - [DeepSeek V4.1 Flash vs V4 Flash 0731: what changed, and what it costs](https://lu-labs.ai/blog/deepseek-v4-1-flash-vs-v4-flash): Two models called Flash, six point seven times apart in price. One reads images and reasons every turn, the other is the cheap workhorse. Which one for which job. - [Flux 2 Dev vs Flux Dev vs Flux Schnell](https://lu-labs.ai/blog/flux-2-dev-vs-flux-dev-vs-flux-schnell): Flux 2 Dev, Flux Dev and Flux Schnell side by side: what each one is for, which is four times cheaper, and which owns the masked edit lane. - [How to Generate Flux 2 Images Online](https://lu-labs.ai/blog/how-to-generate-flux-2-images-online): How to generate Flux 2 images online, in a browser tab, with the credit maths spelled out: what a draft costs, and what a finished frame costs. - [How to run Qwen 3.8 without a GPU](https://lu-labs.ai/blog/how-to-run-qwen-3-8-without-a-gpu): How to run Qwen 3.8 without a GPU: both flagships are trillion-parameter models, so here is the browser route from five euros, and what you can download. - [How to use Kimi K3 online, step by step](https://lu-labs.ai/blog/how-to-use-kimi-k3-online): How to use Kimi K3 online: the browser route, the OpenAI-compatible endpoint for your own code, and how to keep 2.8 trillion parameters off your credits. - [How to Use LU Labs Cloud on a Mac](https://lu-labs.ai/blog/how-to-use-lu-labs-cloud-on-a-mac): How to use LU Labs Cloud on a Mac: the browser Studio set up properly, for the models a Mac cannot hold, plus the API key trick for your editor. - [Kimi K3 vs DeepSeek V4: which one, and what for](https://lu-labs.ai/blog/kimi-k3-vs-deepseek-v4): Kimi K3 vs DeepSeek V4: one reads images, the other costs a fraction per token. Every plan carries both, so the choice is price against the work. - [Qwen 3.8 Max vs A95B: which one, and what for](https://lu-labs.ai/blog/qwen-3-8-max-vs-a95b-explained): Qwen 3.8 Max vs A95B: two flagships in the same picker. The real difference is who controls the reasoning, and what a single turn ends up costing. - [What a Mac Can Run Locally in 2026](https://lu-labs.ai/blog/what-a-mac-can-run-locally-in-2026): What a Mac can run locally in 2026: unified memory decides it. What each memory tier holds, the two walls a laptop cannot climb, and the usual split. - [GLM 5.3 and GLM 5.3 Flash Are Live in LU Cloud](https://lu-labs.ai/blog/glm-5-3-and-glm-5-3-flash-live-in-lu-labs-cloud): GLM 5.3 and GLM 5.3 Flash are in the LU Cloud catalog, on every plan and every credit pack. Both think before answering, hence the new Effort button. - [GLM-5.3 vs GLM-5.2: Same Model, New Training](https://lu-labs.ai/blog/glm-5-3-vs-glm-5-2): GLM-5.3 vs GLM-5.2: we diffed both config files. Every architecture field matches, so 5.3 is 5.2 with new post-training and a stricter licence. - [How to Run GLM-5.3 on Your Own Computer](https://lu-labs.ai/blog/how-to-run-glm-5-3-locally): How to run GLM-5.3 on your own computer: what the 754B flagship and the 321B Flash need in disk and memory, and which one your machine can hold. - [An opencode Alternative That Cost 40 Percent Less](https://lu-labs.ai/blog/opencode-alternative-cost-benchmark): An opencode alternative, measured: the same bug fix, model and prices, run three times through opencode and once through the LU Labs coding agent. - [How to run Qwen 3.8 27B locally, on a PC or a Mac](https://lu-labs.ai/blog/how-to-run-qwen-3-8-27b-locally): Run Qwen 3.8 27B locally: real file sizes for every quant, what it needs in VRAM or unified memory, the Mac route through MLX, and the setting that decides it. - [Kimi K3 and Ling 3.0 Flash Are Live in LU Cloud](https://lu-labs.ai/blog/kimi-k3-and-ling-3-0-flash-live-in-lu-labs-cloud): Kimi K3 and Ling 3.0 Flash are in the LU Cloud catalog: what each one is for, what it costs in credits, and how the thinking toggle keeps K3 affordable. - [Kimi K2.6 in LU Cloud: When to Pick It Over K3](https://lu-labs.ai/blog/kimi-k2-6-agent-workhorse-guide): Kimi K2.6 is the cheaper Kimi, and for long agent runs it is often the right one. What it is good at, where K3 earns its price, and how to use each. - [Kimi K3 explained, from release date to price](https://lu-labs.ai/blog/kimi-k3-explained-and-lu-labs-cloud): Kimi K3 in full: announced 16 July 2026, weights on 27 July, 2.8 trillion parameters, a 1M token context, its own licence, and what it draws in credits. - [Best Open-Weight LLMs to Run at Home in 2026](https://lu-labs.ai/blog/best-open-weight-llms-2026): The open-weight models worth downloading in September 2026: parameters, context windows and licences from the model cards, plus the four no desk can hold. - [Generate AI Images on Your Own Computer](https://lu-labs.ai/blog/generate-ai-images-on-your-own-computer): How to generate AI images on your own computer: free, offline and unlimited. What each model needs in VRAM, which are worth the download, and where a Mac fits. - [How Much RAM Do You Need for Local AI?](https://lu-labs.ai/blog/how-much-ram-do-you-need-for-local-ai): Practical RAM guidelines for local AI in 2026: what runs on 8, 16, 32 and 64 GB, how quantization shrinks models, and how to pick a model that fits. - [How to Run AI Models Locally on Your Mac](https://lu-labs.ai/blog/how-to-run-ai-models-locally-on-your-mac): Step-by-step 2026 guide to running AI models locally on your Mac: which Macs can handle it, how model sizes work, and the easiest apps to get started. - [Introducing LU Labs: A Free Local AI Studio](https://lu-labs.ai/blog/introducing-lu-labs): LU Labs is a free desktop app for Windows, macOS and Linux: a private AI studio for chat, coding agents, image and video generation, all running locally. - [Local AI Video Generation in 2026](https://lu-labs.ai/blog/local-ai-video-generation-2026): An honest look at local AI video generation in 2026: the five open models that run on home hardware, what each costs in VRAM and disk, and where the cloud wins. - [Local AI vs Cloud AI: What Stays Private?](https://lu-labs.ai/blog/local-ai-vs-cloud-ai-privacy): A plain-English comparison of what cloud AI services can see and keep versus what stays on your machine with local AI, and how to choose between them. - [Ollama vs LM Studio, compared in September 2026](https://lu-labs.ai/blog/ollama-vs-lm-studio-vs-lu-labs): Ollama vs LM Studio in September 2026: licences, platforms, model formats, GPU backends, tool calling, headless servers, and who should pick which one. - [What Is an AI Agent, and Why Run One Locally?](https://lu-labs.ai/blog/what-is-an-ai-agent-and-why-run-one-locally): What an AI agent actually is, how it differs from a chatbot, and why running one locally keeps your files and code private. Plus how to try one for free. ## Open source The desktop studio is free and open source, runs entirely on your own hardware and needs no account. LU Cloud is the optional hosted tier for the moments hardware is the bottleneck: heavier models, faster video, phone access. Cloud runs open-weight models only.