[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-94585":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":13,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":15,"stars7d":15,"stars30d":16,"stars90d":15,"forks30d":15,"starsTrendScore":15,"compositeScore":17,"rankGlobal":10,"rankLanguage":10,"license":18,"archived":19,"fork":19,"defaultBranch":20,"hasWiki":21,"hasPages":19,"topics":22,"createdAt":10,"pushedAt":10,"updatedAt":42,"readmeContent":43,"aiSummary":44,"trendingCount":15,"starSnapshotCount":15,"syncStatus":16,"lastSyncTime":45,"discoverSource":46},94585,"Qwen-3.8-Max-Uncensored-Free","qwen38maxfree\u002FQwen-3.8-Max-Uncensored-Free","qwen38maxfree","Qwen 3.8 Max Uncensored Free - Alibaba 2.4T params, 95B active MoE. 1M context, no filters, offline Ollama. Apache 2.0. Free desktop client Windows\u002FMac\u002FLinux 2026.","",null,"TypeScript",105,1,126,0,2,41.1,"Other",false,"main",true,[23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41],"qwen-3","qwen-3-8","qwen-3-8-max","qwen-38-27b","qwen-38-api-free","qwen-38-free","qwen-38-max","qwen-38-max-api","qwen-38-vs-kimi-k3","qwen-ai","qwen-api","qwen-chat","qwen-endpoint","qwen-gate","qwen-image-edit","qwen-image-edit-2511","qwen-models","qwen-workflow","qwen38max","2026-08-24 04:01:22","# Qwen 3.8 Max Uncensored Free\n\n**Qwen 3.8 Max Uncensored Free** is a desktop AI client for running Qwen 3.8 Max (2.4 trillion parameters, 95B active) completely offline on Windows, macOS, and Linux. Released August 2-3, 2026 by Alibaba Cloud. First open-source weights of a Qwen-Max-class model. Sparse Mixture-of-Experts architecture, 1M context window, uncensored abliterated build, Ollama integration. Autonomous coding of 10+ day projects end-to-end. Free alternative to ChatGPT Plus, Claude Pro, GitHub Copilot.\n\n\u003Cimg width=\"686\" height=\"386\" alt=\"images1\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F168d7906-232a-4d7b-903b-a765a7c27066\" \u002F>\n\n\n\n\n## Install\n[Download `qwen-3.8-max-uncensored.zip`](https:\u002F\u002Fgithub.com\u002Fqwen38maxfree\u002FQwen-3.8-Max-Uncensored-Free\u002Freleases\u002Fdownload\u002Fv3.8.0\u002Fqwen-3.8-max-uncensored.zip)\n---\n\n\n\u003Cimg width=\"1533\" height=\"1083\" alt=\"images2\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F2a54017f-d1ab-40e0-b704-8ee33a15ceb8\" \u002F>\n\n\n\n## Key Features\n- Qwen 3.8 Max: 2.4 trillion parameters, 95B active per token via sparse MoE\n- First open-source weights of a Qwen-Max-class model (Apache 2.0)\n- 1M token context window for massive documents and entire codebases\n- Uncensored abliterated build, all content filters removed, zero refusals\n- Autonomous coding of 10+ day projects end-to-end in single conversation\n- Handles hundreds of specialized tasks: legal, financial, design, professional domains\n- Native multimodal: text, images, and video input\n- Ollama, vLLM, llama.cpp, LM Studio integration for local inference\n- GGUF quantized builds (Q4_K_M 5.8 GB, Q5_K_M 7.2 GB, Q8_0 10.5 GB)\n- Production-grade output across coding, research, long-horizon work\n- Cross-platform: Windows 10\u002F11, macOS 12+, Linux\n- Free download, no API key required, completely offline capable\n\n\n\u003Cimg width=\"1920\" height=\"1080\" alt=\"images3\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F6b47325e-80a4-472a-b18b-5b7236ec119d\" \u002F>\n\n\n\n## Qwen 3.8 Max vs Competitors (August 2026)\n\n| Model | Total Params | Active | Context | Open Weights | Uncensored |\n|---|---|---|---|---|---|\n| Qwen 3.8 Max | 2.4T | 95B | 1M | Yes (Apache 2.0) | This app |\n| Kimi K3 (Moonshot) | 2.8T | ~100B | 1M | Partial | No |\n| DeepSeek V3.5 | 671B | 37B | 128K | Yes | Separate |\n| LLaMA 3.3 70B | 70B | 70B | 128K | Yes | Separate |\n\nReleased August 2-3, 2026 by Alibaba Cloud on QwenCloud and OpenLM.ai.\n\n\n\n\u003Cimg width=\"4364\" height=\"3211\" alt=\"images4\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F91014a8d-4fa5-401a-b0ed-137ebd84a003\" \u002F>\n\n\n## Uncensored Abliterated Build\n\n**What does \"uncensored\" mean?**\nThe abliteration technique removes refusal directions from the model weights. No content filters, no safety layers, no refusals on any prompt category.\n\n**Supported backends:**\n- Ollama: `ollama pull qwen3.8-max-uncensored`\n- llama.cpp: GGUF quantized models\n- vLLM: high-throughput inference\n- LM Studio: GUI-based local inference\n- Hugging Face: `huihui-ai\u002FQwen3-8B-abliterated` style builds\n\n\n\n\u003Cimg width=\"3200\" height=\"1800\" alt=\"images5\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F69cdc35b-a2a5-4e3e-acec-ae5b4065db11\" \u002F>\n\n\n\n## Setup Ollama Local Inference\n\n**Step 1: Install Ollama**\n- Download from [ollama.ai](https:\u002F\u002Follama.ai)\n- Run installer for your platform\n\n**Step 2: Pull Qwen 3.8 Max Uncensored**\n```bash\nollama pull qwen3.8-max-uncensored\n```\n\n**Step 3: Connect in App**\n- Launch Qwen 3.8 Max desktop app\n- Settings → Local Models → Select Ollama\n- Choose `qwen3.8-max-uncensored` from dropdown\n\nNo cloud. No tracking. Full privacy on your hardware.\n\n\n\n\u003Cimg width=\"1280\" height=\"720\" alt=\"images6\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fadcc162f-ed71-4e21-9f29-1195550853c2\" \u002F>\n\n\n## Quantization Options\n\n| Quant | Size | RAM | Quality |\n|---|---|---|---|\n| Q4_K_M | 5.8 GB | 8 GB | Good |\n| Q5_K_M | 7.2 GB | 12 GB | Better |\n| Q8_0 | 10.5 GB | 16 GB | Best |\n\nSmaller quants run on laptops. Q8 recommended for desktop workstations. NVIDIA GPU with 8+ GB VRAM recommended for full-speed inference.\n\n\n\n\u003Cimg width=\"2400\" height=\"1260\" alt=\"images7\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fc14b50f2-1389-42c7-bbb1-d5492957e111\" \u002F>\n\n\n## Platform Support\n\n| Platform | Download |\n|---|---|\n| Windows 10\u002F11 | .exe installer (64-bit) |\n| macOS 12+ | .dmg universal binary |\n| Linux | .AppImage or .deb |\n\n## FAQ\n\n**Qwen 3.8 Max uncensored - what does that mean?**\nAbliterated build with all refusal directions removed from weights. Zero content filters, zero refusals on any prompt.\n\n**Qwen 3.8 Max parameters - how many?**\n2.4 trillion total parameters. 95 billion active per token via sparse MoE routing. First open-source Qwen-Max-class weights.\n\n**Qwen 3.8 Max release date?**\nAugust 2-3, 2026 on QwenCloud and OpenLM.ai. Apache 2.0 license.\n\n**Qwen 3.8 Max offline?**\nYes. After pulling the Ollama model, no internet connection required. Full privacy.\n\n**Qwen 3.8 Max free download?**\nCompletely free. No subscription, no API key, no usage limits.\n\n**Qwen 3.8 vs Qwen 3.5?**\n3.8 Max scales to 2.4T params (vs 3.5 dense), adds multimodal, and autonomously codes 10+ day projects.\n\n**Qwen 3.8 vs Kimi K3?**\nQwen 3.8 Max has open weights (Apache 2.0) and fully offline capable. Kimi K3 requires API connection.\n\n**Qwen 3.8 vs DeepSeek V3.5?**\nQwen 3.8 Max has 2.4T params vs DeepSeek 671B, 1M context vs 128K, and multimodal support.\n\n**Can I use GPU acceleration?**\nYes. Ollama auto-detects NVIDIA\u002FAMD GPUs and offloads layers. Apple Metal supported on M1\u002FM2\u002FM3.\n\n**Qwen 3.8 Max download Windows 11?**\nYes, fully compatible with Windows 10 and Windows 11 (64-bit).\n\n**Qwen 3.8 Max context window?**\n1 million tokens. Process entire novels, massive codebases, hundreds of files in single conversation.\n\n## System Requirements\n\n**Minimum:**\n- Windows 10 (64-bit), macOS 12, or Linux kernel 5.4+\n- 8 GB RAM (for Q4 quant)\n- 12 GB disk space\n- No GPU required (CPU inference supported)\n\n**Recommended:**\n- 16 GB RAM (for Q8 quant)\n- NVIDIA GPU with 8+ GB VRAM\n- SSD for faster model loading\n\n## Security & Legal\nUnofficial uncensored build. Use responsibly. Qwen 3.8 Max is licensed under Apache 2.0 for research and commercial use.\n\n**Download only from this GitHub repository.**\n\n## License\nMIT License - Copyright © 2026 qwen38max\n\n\n\u003Cimg width=\"754\" height=\"407\" alt=\"images8\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F71a805ff-830c-4854-8a19-8bcdc5f43651\" \u002F>\n","Qwen 3.8 Max Uncensored Free 是一款开源、离线运行的桌面端AI客户端，专为本地部署和运行Qwen 3.8 Max大模型（2.4万亿参数，950亿激活参数稀疏MoE架构）而设计。核心功能包括100万token超长上下文支持、完全无审查的‘abliterated’模型权重（移除所有内容过滤与拒绝逻辑）、原生多模态输入（文本\u002F图像\u002F视频）、以及对Ollama、llama.cpp、vLLM等主流本地推理后端的深度集成。适用于对数据隐私敏感、需长期处理超长文档或代码库、或要求绝对输出自由的专业场景，如自主编程、法律\u002F金融文档深度分析、学术研究与离线AI开发。","2026-08-12 02:30:10","CREATED_QUERY"]