Qwen 3.8 vs GPT 5.6 vs Fable 5 vs Kimi K3: Vibe-Coding Showdown

Key Takeaways

• Qwen 3.8 is a 2.4-trillion-parameter model that its makers position as "second only to Fable 5" and are preparing to release as open weights.
• Qwen led on graphics and creativity across the ~23 vibe-coded builds — winning the Outrun, GTA-style, Dragonfly, and landing-page tests — but consistently shipped worse controls, camera angles, and gameplay feel.
• Fable 5 remained the reviewer's overall favorite for UI polish and the Remotion video trailer; it struggled on some color-heavy and Skyrim-style builds.
• Kimi K3 placed a close second, leading on physics (a movable cloth simulation), fireworks, and Dragon Realm, and is credited with topping front-end coding benchmarks.
• GPT 5.6 Soul landed mid-pack — solid camera and controls on some RPG builds, but the weakest Skyrim example and a very basic, AI-looking landing page.
• Practical friction with Qwen: Qwen.com is region-locked (unreachable from the UK) and the international API wouldn't connect, so testing ran through Koda/QD. The reviewer's verdict is "test it yourself" — other reviewers on X rated Qwen far lower.