01What Qwen 3.8 vs Fable 5 Comparison Shows
Qwen 3.8 vs Fable 5 is a head-to-head test of two leading AI models for creating playable games from a single prompt. In this episode, Julian Goldie runs 23 different builds side-by-side to see which model produces the best results. Qwen 3.8 is a massive 2.4 trillion parameter model that just launched and is going open-source soon.
Fable 5 is currently Julian's favorite model. The comparison covers game quality, graphics, controls, and fun factor. Julian tests everything from flight simulators to RPGs, racing games, and even a Minecraft-style build.
The goal is to see if Qwen 3.8 can match the quality that Fable 5 is known for.
02Main Features of Qwen 3.8 and the Test Setup
Qwen 3.8 comes with a huge 2.4 trillion parameter count. According to the company, it is one of the most powerful AI models available today. They claim it is second only to Fable 5 among frontier models.
Julian notes that Qwen 3.8 shows a big step up from the previous Qwen 3.7 version.
- 23 different game builds tested side-by-side
- Flight simulators with detailed graphics and smooth gameplay
- Open-world games like Skyrim and GTA styles
- Physics tests including cloth simulation and fluid dynamics
- Web design and video generation comparisons
Julian also mentions he cannot access Qwen.com from the UK due to regional restrictions. He used Coda to test Qwen 3.8 instead of the international API, which would not work for him.
03How to Get Started Testing These AI Models
If you want to try these models yourself, Julian explains how he set up his tests. He created a benchmark called GoldieBench to run all 23 tests. He plans to expand it to 50 tests once everything is completed.
- Pick a set of game prompts you want to test, like a flight simulator or racing game.
- Run the same prompt on each AI model you want to compare.
- Play the resulting game and check graphics, controls, and fun factor.
- Score each model on quality, smoothness, and detail.
- Repeat across multiple games to find the best overall model.
Julian notes that running every test takes about a full day. The tests run in the background while you do other things. He recommends testing for yourself instead of trusting other people's opinions.
04Walking Through the Game Build Examples
Julian goes through many specific examples in the episode. In the Outrun-style game, Qwen 3.8 created the best output by far. It looked the nicest, was the most fun, and had the most interesting design.
GPT 5.6 Sol came out worst on that test.
For the Skyrim example, Fable 5 was really nice with smooth quality and little details like a compass at the top. Kimi K3 came second. Qwen 3.8 had crazy gameplay but the graphics were not quite as nice.
The controls were also weird, with moving forward taking you backward in the GPT 5.6 version.
In the GTA-style game, Qwen 3.8 absolutely crushed the test. The quality, detail, and fun factor were excellent. Fable 5 had less detail and was not as fun.
Kimi K3 did not even create a car, and GPT 5.6 required you to enter the car manually.
05Test Results and Comparisons Across Categories
Results were a mixed bag across the 23 tests. Qwen 3.8 won several categories including the Outrun game, GTA-style game, Dragonflight, and a flying game. Fable 5 won the flight simulator and some other tests.
Kimi K3 absolutely crushed the Dragon Realm example with nice looks, good ambience, and solid UI.
- Physics tests: Kimi K3 clearly won, allowing cloth interaction that other models did not support
- Fireworks example: Qwen 3.8 came in last, Kimi K3 won again
- Landing page design: Qwen 3.8 created by far the nicest design, breaking typical AI web design boundaries
- Video generation: Fable 5 did the best, followed closely by Qwen 3.8
- Fluid dynamics: Qwen 3.8 did not even come close, Kimi K3 performed best
Overall, Julian ranks Fable 5 as his favorite, with Kimi K3 second. He says Qwen 3.8 is up there but not quite on par with Kimi K3.
06Tips, Limits, and Mistakes to Avoid With Qwen 3.8
Julian shares several important tips for using Qwen 3.8 effectively. First, expect to do some back-and-forth prompting to get good outputs. It takes effort to get the quality results shown in the tests.
- Regional access issues: Qwen.com may not be accessible in all countries; Julian could not access it from the UK
- Control bugs: Some games had problems with movement, throttle, and camera controls
- Inconsistent quality: Qwen 3.8 can create awesome stuff on some tests but fails on others
- UI preference: Fable 5 generally has better UI and ambience than Qwen 3.8
- Lighting issues: The Doom-style game had very bad lighting, making it hard to see
Julian warns that other people on X have posted negative comparisons. For example, a user named Omed the vibe coder said Qwen was nowhere near the same level. Julian recommends always testing for yourself rather than relying on others' opinions.
07Who Should Use Qwen 3.8 and Why It Matters
Qwen 3.8 is for anyone building games or creative projects with AI. It matters because it proves that open-source models can compete with the best commercial ones. The 2.4 trillion parameter scale brings serious capability to game generation.
Julian says Qwen 3.8 holds its own against Fable 5. It can build really amazing stuff and creates smooth experiences. However, he still prefers Fable 5 for its UI quality and overall feel.
Kimi K3 remains his second choice overall.
For AI automation and scaling businesses, Julian recommends joining the AIPUF community. It includes master classes on all these models, weekly coaching calls, and daily updates on new developments. You also get access to his best trainings and can ask questions that he personally answers every day.
08Final Verdict on Qwen 3.8 vs Fable 5
Julian's final verdict is clear: Fable 5 is still his favorite model today. He would not say Qwen 3.8 is better than Fable 5, nor would he say it is on par with Kimi K3. However, he believes Qwen is up there with the top models.
The key takeaway is that there is no single best AI for everything. Sometimes you prompt different models exactly the same way and get very different outputs. Qwen 3.8 wins on some tests, loses on others, and produces mixed results throughout.
For the landing page test, Qwen 3.8 by far created the nicest design. It felt less like AI and more like professional web design. But for physics and cloth interaction, Kimi K3 was clearly superior.
Julian encourages everyone to test these models for themselves and see which ones work best for their specific needs.
