Qwen 3.8 Next First Test | Coding, Game Dev, Frontend | 2.4T Parameter Open Weight model by Alibaba

2026-07-19 · Source: Venelin Valkov · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, medium

Summary

Alibaba's Qwen 3.8 Max Preview, a large language model with approximately 2.4 trillion parameters, was tested for its coding capabilities in game development and frontend design. Despite Alibaba's claim of its compatibility with leading frontier AI models, second only to Fable 5, the model demonstrated inconsistent performance. Initial tests, such as generating a Snake game, resulted in the model getting stuck in a thinking loop for over seven minutes without producing code. While it successfully generated HTML for an embodied gravity simulation and a boating game, the output speed was slow (around 20-25 tokens/second), and the quality, physics, and design were deemed inferior to models like Fable 5, Kimik A3, and GM 5.2. A final test for an ML/AI engineer website also yielded a "pretty bad" design compared to competitors, leading to an overall unimpressed assessment by the reviewer.

Key takeaway

For AI Engineers evaluating new open-weight models for coding and frontend development, Qwen 3.8 Max Preview, despite its 2.4 trillion parameters, currently underperforms leading frontier models. You should exercise caution and conduct extensive testing, especially for complex tasks involving game logic or intricate UI design, as it exhibited issues like getting stuck in thinking loops and producing suboptimal code. Consider alternatives for production-grade coding applications.

Key insights

Alibaba's 2.4T parameter Qwen 3.8 Max Preview model shows inconsistent coding performance, often struggling with complex tasks and exhibiting slow "thinking."

Principles

• Large parameter count does not guarantee superior performance.
• Model "thinking" process can indicate internal issues like loops.
• Coding interpreters can aid complex code generation.

Method

The author tested Qwen 3.8 Max Preview via its chat interface using heavy prompts for game development (Snake, boating) and frontend design (gravity simulation, engineer website), observing output quality, speed, and internal "thinking" processes.

In practice

• Test large language models with complex, multi-component coding tasks.
• Monitor model "thinking" logs for signs of internal processing issues.
• Compare model output against established frontier models for specific tasks.

Topics

• Qwen 3.8 Max Preview
• Large Language Models
• Code Generation
• Frontend Development
• Game Development
• Model Benchmarking

Best for: AI Engineer, Machine Learning Engineer, AI Scientist

Related on AIssential

• 😸 Alibaba’s 2.4T Qwen joins the AI race — Alibaba's 2.4T Qwen3.8-Max model, promised as open-weight, intensifies the competition between proprietary and accessible frontier AI.
• Qwen 3.8 Max IS INSANE! Second To Fable? New Open Model King? (Fully Tested) — Qwen 3.8 Max is a powerful multimodal model excelling in complex code generation and visual rendering, despite not matching top-tier benchmarks.
• Qwen3.8-Max-Preview Is Live on AIHubMix-90% Off for Launch Week — Qwen3.8-Max-Preview, a 2.4T-parameter model, is available via an OpenAI-compatible API with a 90% launch discount.
• Alibaba's Qwen takes on Kimi K3 with open-weight Qwen 3.8, says model is "second only to Fable 5" — Alibaba's Qwen 3.8, a 2.4T parameter multimodal model, challenges Kimi K3 with open weights and claims near-Fable 5 performance.
• Alibaba unveils 2.4T-parameter Qwen3.8 AI model — Alibaba's 2.4T-parameter Qwen3.8 model is launching with a new open-weight strategy, despite unverified performance claims.
• Qwen 3.8 Released : Better than Kimi K3? — Alibaba's Qwen 3.8, a 2.4 trillion parameter model, is set for an open-weight release, emphasizing deployment flexibility.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Venelin Valkov.