Verdict
Built from the measurements below
- Qwen3.8 Max (0902) is more accurate: 75.0% against 32.5% on the same 80 questions, a 42.5-point gap that is statistically significant.
- Qwen3.8 Max (0902) is 1.2× cheaper per question as billed ($0.011 vs $0.013), reasoning tokens included.
- Qwen3.5 397B A17B generates text 5.2× faster: 240 tokens per second against 46.
- Qwen3.5 397B A17B starts answering sooner: 0.6 s to the first token, against 1.2 s for Qwen3.8 Max (0902).
- Qwen3.8 Max (0902) accepts more context: 1M tokens against 262k.
Which should you choose?
Choose Qwen3.5 397B A17B for faster streaming, a quicker first token, open weights you can run yourself and more hosts to choose from on OpenRouter (10 against 1). Choose Qwen3.8 Max (0902) for the most accurate answers of the two, the lower cost per correct answer ($0.015 against $0.041), shorter outputs (about 1,722 tokens per question against 4,010) and a 1M-token context window. The widest subject gap is in engineering: Qwen3.8 Max (0902) answered 9 of 15 correctly, Qwen3.5 397B A17B 0 of 15. With so few questions per subject, treat this as a lead to test on your own prompts, not a result.
Accuracy gap -42.5 points (z = -5.39). We call a winner only when z is at least 1.96, the usual 95% threshold. Speed, latency and price need a 10% margin.