Qwen4-Max hands-on: massive gains over Qwen3.8-Max, even in embedded dev
karminski3 · x · 2026-11-20
Bilibili creator karminski-牙医 published a full hands-on review of the Qwen4-Max release. Having tested the model during internal beta, he re-ran the same one-shot ray-tracing prompt after launch and found a dramatic gap — the final version is far better — so he re-did all his tests.
The review covers:
- An embedded program with AI-generated components (embedded dev scenario)
- A leaderboard tool that fetches HuggingFace open-weight model rankings
- A 3-render-2 AI calendar
Verdict: huge improvement over Qwen3.8-Max, with especially notable gains in frontend and agent capabilities; backend agentic coding "seems promising" — the video examines whether it can serve as a daily-driver coding model.
Related event: Hands-on Tests Show Qwen4-Max Makes Big Gains in Frontend and Agent Skills(2 posts)→
More from Models
- Google AI Mode now enforces usage limits, users surprised by rate-limit popup — elvano124 · 2026-11-22
- Sol 6.2 now refuses reverse engineering that Sol 6.1 happily did for a month — cardak98 · 2026-11-22
- Anthropic's record ~$81B IPO, GPT-6.2 pricing, Qwen4-Max and Copilot Autopilot: week in review — ksraj1001 · 2026-11-22
- Reddit rant: Frontier AI labs are training models to imitate lazy engineers — RossPeili · 2026-11-22
- MBZUAI open-sources PAN 2, a 29B-parameter general-purpose world model — LinusEkenstam · 2026-11-22
- Inkling Mini turns out surprisingly fluent in Russian, sparking data recipe questions — teortaxesTex · 2026-11-22