011. What Qwen 3.8 27B Is and How It Works

Qwen 3.8 27B is a free, open source AI model you can download and run on your own computer. It’s the local version of Alibaba’s Qwen 3.8 series. The “27B” means it has 27 billion parameters, which makes it powerful but still possible to run on consumer hardware.

This model is open weights, so anyone can use it without paying. It works by running entirely on your machine, not on a cloud server. That gives you privacy and lets you use AI even when you have no internet.

In this episode, Julian and his co-host explain that Qwen 3.8 27B is one of the best free local models you can actually use today. It’s not a toy – it can handle real tasks.

022. Main Features of Qwen 3.8 27B

Qwen 3.8 27B comes with several important features that make it stand out among local AI models.

  • Free and open source: You pay nothing and can inspect the model weights. No subscriptions or API costs.
  • Private: Everything runs offline on your computer. No data leaves your machine.
  • Frontier-level performance: It performs at a level similar to Claude Opus 4.6, which was considered the best model just six months ago.
  • Easy to set up: You can install it using free tools like Ollama or LM Studio. Ollama even updated its Qwen 3.8 support just 16 hours before this episode.
  • Optimized for Mac (MLX): The model can run on Apple Silicon Macs using the MLX framework, which makes it smoother on those machines.

033. How to Set Up Qwen 3.8 27B

Setting up Qwen 3.8 27B is straightforward. The easiest method uses Ollama, but you can also use LM Studio. Follow these steps:

  1. Check your hardware. You need a computer with a powerful GPU – for example, an RTX graphics card or a Mac with enough RAM. The model file is about 18 GB. Laptops like a MacBook Pro or Mac Mini might struggle.
  2. Install Ollama. Go to the Ollama website and download the free app. Ollama works on Windows, macOS, and Linux.
  3. Get the model. Open your terminal and run the command ollama pull qwen:3.8-27b. Ollama will download the model automatically. The episode says Ollama was updated 16 hours ago with the new Qwen 3.8 support.
  4. Alternatively, use LM Studio. Download LM Studio (a free app). Inside LM Studio, you can browse and download Qwen 3.8 27B directly, then run it with a simple chat interface.
  5. Verify it works. Type a test prompt. The model should respond in real time. If it’s too slow, your hardware might not be powerful enough.

044. How to Use Qwen 3.8 27B in Practice

Once Qwen 3.8 27B is installed, you can use it for many AI tasks. The episode mentions two specific ways to use it beyond basic chat.

Use with Claude Code. You can connect Qwen 3.8 27B to Claude Code, which lets you run AI‑assisted coding tasks locally. This gives you a free coding assistant that works offline.

Use as a brain for open code tools. Julian and his co-host also mention using Qwen 3.8 27B as the underlying model for tools like Open Code Hermes or Open Core. In that setup, Qwen 3.8 27B answers questions and generates code, while the tool handles the interface.

You can also simply chat with it through Ollama’s terminal or LM Studio’s chat window. Because it’s a large model, responses are smart and detailed – similar to what you would get from a cloud service like Claude or ChatGPT.

055. Results, Tests, and Comparisons

The episode shares a key comparison: Qwen 3.8 27B performs at the same level as Claude Opus 4.6. That model was considered the best around six months ago. So this free local model is not far behind current frontier models.

Julian and his co-host note that the gap between local AI and cloud AI is closing fast. A year or two ago, local models were one or two years behind. Now that gap is shrinking.

Models like Qwen 3.8 27B are proof.

However, they also share a warning from their own experience: some local models they tested on the Goldie Bench leaderboard were terrible. They were slow, produced low‑quality output, and sometimes failed to create anything useful. Qwen 3.8 27B is different – it actually works well if you have the right setup.

066. Tips, Limits, and Mistakes to Avoid

Using Qwen 3.8 27B locally has real limits. Here are the key tips and mistakes to watch out for.

  • You need strong hardware. The model is 18 GB. Even on a Mac Studio, it may not run smoothly. A Mac Mini or MacBook Pro probably can’t handle it well. A PC with an RTX GPU or a DGX Spark is a better bet.
  • Speed can be disappointing. In the episode, Julian warns that free self‑hosted models often sound great until you actually run them and discover how slow they are. Qwen 3.8 27B is faster than many, but on weak hardware it will be unusable.
  • Use Ollama for simplicity. The co‑host recommends Ollama as the easiest method. It’s just one command in the terminal. LM Studio is also fine but not his personal preference.
  • Mac users should try MLX. If you are on a Mac, look for an MLX‑optimized version of Qwen 3.8 27B. That can make it run smoother than the standard version.
  • Don’t expect top performance on laptops. Unless you have a high‑end gaming laptop, your machine may struggle. The episode mentions that running this model on a Mac Mini or MacBook Pro is probably not going to work well.

077. Who Qwen 3.8 27B Is For and Why It Matters

Qwen 3.8 27B is for people who want private, free AI without an internet connection. That includes developers who need a local coding assistant, privacy‑conscious users, and anyone who works in places with no Wi‑Fi – like on an airplane.

It also matters because it shows how fast local AI is improving. The episode points out that frontier‑level models are now being released more frequently, and the quality gap with cloud models is shrinking. In a couple of years, local AI may be just as good as the best cloud services.

Right now, Qwen 3.8 27B gives you near‑frontier AI for zero cost, as long as you have the right hardware. It’s a glimpse of a future where powerful AI runs on your own computer, not someone else’s server.