16 November 2026 · 2 min read
I Tested Qwen 4 Max: Alibaba’s 2.4T AI Model vs Claude (10x Cheaper?)
I put Alibaba's new Qwen 4 Max through real coding tests—building games, 3D environments, and tools—and compared it directly with Claude. Here's how the 2.4T parameter model performed and what it cost.
Alibaba just released Qwen 4 Max, their most capable model yet. It packs 2.4 trillion parameters, a 1 million token context window, and pricing that undercuts many frontier models at $1.50 per million input tokens and $4.50 per million output tokens. The company claims it beats Claude Opus 5.5 and GPT-6.1 Sol at agentic coding. In my latest video, I put those claims to the test with my own coding challenges.
What Qwen 4 Max actually is
Qwen 4 Max is a sparse mixture-of-experts model. Only 71 billion parameters are active per token, which keeps inference efficient despite the enormous total size. It accepts text and images (native vision), supports up to 262K output tokens, and lets you dial reasoning effort up or down. The weights are closed, but Alibaba has promised smaller Qwen 4 models as open source.
Benchmarks: where it wins and where it loses
Alibaba’s launch numbers show strong results on a few benchmarks. On DeepSWE v1.1 it scores 75.7, edging out Kimi K3.1 (75.5), GPT-6.1 Sol (75.2) and Opus 5.5 (74.2). On Agents’ Last Exam v1.1 it reaches 46.7, ahead of Opus 5.5 at 43.4.
But the picture isn’t one-sided. On Terminal-Bench 4.0, Qwen 4 Max scores 48.5 while Kimi K3.1 hits 67.2. On ProgramBench, it manages 31.5 against Opus 5.5’s 65.0. The independent Vals AI Index places it at #11 out of 38 models—the top Chinese model, but not the overall leader. I go through these numbers honestly in the video so you can see where the model genuinely excels and where it still falls short.
Real coding tests inside VS Code
I didn’t stop at benchmarks. I loaded Qwen 4 Max into Kilo Code in VS Code and built five projects from scratch:
- A racing game from a single screenshot
- A 3D Three.js environment
- An interior design tool
- A 3D Rubik’s Cube simulator
- A 2D driving game
For several of these, I ran the exact same prompts through Claude Fable 5.2 and compared the outputs side by side. I also tracked the token usage and cost for each task, so you can see what “10x cheaper” actually looks like in practice.
Who this video is for
If you’re a developer or founder who builds with AI coding tools, this video gives you a practical, no-hype look at a new model that’s suddenly very competitive on price. You’ll see what Qwen 4 Max can generate, where it stumbles, and whether it might replace your current go-to for certain kinds of work.
Watch the full video above for the side-by-side builds, the cost breakdown, and my honest take on whether Qwen 4 Max earns a spot in your toolkit.
Sources
- qwen 4 max
- alibaba ai
- coding benchmark
- claude comparison
- ai coding
Watch the full video on YouTube
Written with AI from the video above.
Learn to build with AI
Hands-on tutorials on Cursor, Claude Code, AI agents and more.
