In this video, I'll be testing the new Qwen 3.8 Max Preview, a massive 2.4 trillion parameter model that Qwen has confirmed will be open-weight. I’ll run it through my full KingBench benchmark, compare it against models like Fable 5, Opus 4.8, Kimi K3, GLM-5.2, and GPT-5.6 Sol, and share my real experience using it with Claude Code.

--
Key Takeaways:

🚀 Qwen has announced Qwen 3.8, a massive 2.4 trillion parameter model with confirmed open weights.
⚡ Qwen 3.8 Max Preview is already available through the Token Plan, Qoder, and QoderWork.
📊 On KingBench, Qwen 3.8 Max scored 65 out of 80, landing in second place overall.
🏆 It finished just behind Fable 5 and above Opus 4.8 on the full benchmark leaderboard.
🎯 The model performed extremely well on frontend, Three.js, SVG, game, math, and agentic tasks.
🧠 Qwen 3.8 Max got perfect scores on the bow and arrow game, the hard math problem, and the long-horizon agentic task.
🛠️ In Claude Code, the model felt fast and capable, though it may still struggle a bit with longer-running tasks.
🌍 With open weights confirmed, Qwen 3.8 could become a major model for the open-source AI community.

AICodeKing

Qwen 3.8 Max, a 2.4 trillion parameter open-weight model from Alibaba's Qwen team, has been benchmarked against top closed models. Tested across 8 tasks including 3D rendering, SVG generation, math reasoning, and agentic workflows, it scored 65/80 (81.25%), placing second overall — just behind Fable 5 and above Claude Opus 4. It achieved perfect scores on a bow-and-arrow game, a hard math permutation problem, and a long-horizon fine-tuning task. The preview is already accessible via their token plan and on Coder/Coder Work. One noted limitation is inconsistent behavior on long-running agentic sessions with Claude Code. The open weights are confirmed but not yet released.

Qwen 3.8 Max (Fully Tested): AN ACTUAL OPEN FABLE COMPETITOR!