01What is Qwen 3.8 and How It Works

Qwen 3.8 is a massive AI model from Alibaba with 2.4 trillion parameters. It generates code for games and apps. In this episode, Julian tested it side-by-side with Fable 5 and GPT 5.6.

The model creates playable games, simulations, and videos from simple prompts. It works by processing natural language and outputting code. You can run it through tools like Coda.

Alibaba claims it is one of the most powerful models available, second only to Fable 5. But real tests show mixed results. Some outputs look great, while others have issues with graphics and usability.

02Main Features and What’s New in Qwen 3.8

Qwen 3.8 brings several new features that stand out in tests.

  • Massive scale: 2.4 trillion parameters, making it one of the largest models.
  • Game generation: Creates 3D games with good graphics and smooth gameplay.
  • Video creation: Produces video content with unique styles, avoiding typical AI look.
  • Timing control: Generates animations with adjustable timing for better flow.

It is a big step up from Qwen 3.7, which was not serious. The new model can build complex projects like open-world RPGs and flight simulators.

However, it has weaknesses in UI design and color choices. Many outputs feel basic or robotic compared to rivals.

03How to Set Up and Access Qwen 3.8

Getting started with Qwen 3.8 is a bit tricky, but you have a few options. Here are the steps Julian showed.

  1. Pick a platform that supports Qwen 3.8, like Coda or Alibaba's token plant.
  2. If you use Alibaba's token plant, you must fill in many details to get access. It is messy.
  3. Check if Qwen is available in your region. For example, UK users cannot access qwen.com.
  4. Use Coda as the easiest way to start building with Qwen 3.8.

Once you have access, you can prompt it to make games or other apps. Remember that region restrictions may block you.

04How to Use Qwen 3.8 with Examples from the Episode

Using Qwen 3.8 means giving clear prompts and testing outputs. In this episode, Julian tested many games side-by-side.

He created a racing game, a fireworks simulation, a black hole, a neon racer, a flight simulator, a video, a Dragon Realm RPG, a GTA-style game, Doom, a dragon flight game, and a Nordic Crypt game.

For most 3D games, Qwen 3.8 produced fun, good-looking results. For example, the neon racer and dragon flight games were smooth and enjoyable.

But some outputs had problems. The Doom game had weird colors and lighting. The Dragon Realm game looked basic with simple graphics.

Tip: Give examples to improve results. Julian noted that Fable 5 might do better with an example.

05Results, Tests, and Comparisons in the Episode

Julian ran many tests to compare Qwen 3.8 with Fable 5 and GPT 5.6. The results were mixed.

Fable 5 vs Qwen 3.8: Fable 5 had cleaner, more elegant designs. For cloth simulation and promo videos, Fable 5 was much nicer. But Qwen 3.8 won on the racing game, neon blaster, and dragon flight game.

GPT 5.6 vs Qwen 3.8: GPT 5.6 had better UI and front ends. In the fireworks test, GPT 5.6 looked nicer. For the flight simulator, GPT 5.6 worked better and allowed takeoff.

But Qwen 3.8 was more fun in games like the neon racer.

Kimi K3 vs Qwen 3.8: Kimi K3 outperformed Qwen 3.8 in graphics, detail, and overall vibe. It topped front-end coding benchmarks.

06Tips, Limits, and Mistakes to Avoid

When using Qwen 3.8, keep these points in mind.

  • Access is hard: It may not be available in your region. Use Coda to get in.
  • UI and graphics can be weak: In many tests, the design looked basic or robotic.
  • Don’t expect perfect results: Some games were unplayable due to color or lighting issues.

Mistakes to avoid: Don’t compare only with benchmarks. Try side-by-side tests to see real performance. Also, don’t assume it is as good as Fable 5 or Kimi K3 just because of marketing.

Julian found that Qwen 3.8 holds its own against frontier models, but it is not the best. It is a good model but not a game-changer.

07Who It Is For and Why It Matters

Qwen 3.8 is for developers and AI enthusiasts who want to build games or apps quickly. It is especially useful if you need a large model that can generate code.

It matters because it offers an open-source alternative to proprietary models like Fable 5 and GPT 5.6. This can reduce costs and increase control.

However, if you want the best graphics and UI, you might prefer Kimi K3 or Fable 5. For big projects, Julian recommends Kimi K3, possibly with Claude Code.

Try Qwen 3.8 if you are curious, but set realistic expectations. It may not blow you away, but it is worth a test.