Qwen3.8 27B vs Qwen3.8 Max (0902): accuracy, speed and price

• Accuracy is too close to call: Qwen3.8 27B scored 63.8% and Qwen3.8 Max (0902) 75.0% on the same 80 questions, a gap within what chance alone could produce.
• Qwen3.8 27B is 2.6× cheaper per question as billed ($0.0043 vs $0.011), reasoning tokens included.
• Output speed is similar (49 vs 46 tokens per second).
• Qwen3.8 27B starts answering sooner: 0.5 s to the first token, against 1.2 s for Qwen3.8 Max (0902).

Which should you choose?

Choose Qwen3.8 27B for the lower cost per correct answer ($0.0067 against $0.015), a quicker first token, open weights you can run yourself and more hosts to choose from on OpenRouter (21 against 1). The widest subject gap is in engineering: Qwen3.8 Max (0902) answered 9 of 15 correctly, Qwen3.8 27B 6 of 15. With so few questions per subject, treat this as a lead to test on your own prompts, not a result.

Accuracy gap -11.3 points (z = -1.54). We call a winner only when z is at least 1.96, the usual 95% threshold. Speed, latency and price need a 10% margin.