Verdict
Built from the measurements below
- Accuracy is too close to call: DeepSeek V4.1 Flash 1017 scored 73.8% and GPT-6 Luna 72.5% on the same 80 questions, a gap within what chance alone could produce.
- GPT-6 Luna is 3.7× cheaper per question as billed ($0.00025 vs $0.00091), reasoning tokens included.
- GPT-6 Luna generates text 2.4× faster: 137 tokens per second against 57.
- DeepSeek V4.1 Flash 1017 starts answering sooner: 0.9 s to the first token, against 1.2 s for GPT-6 Luna.
Which should you choose?
Choose DeepSeek V4.1 Flash 1017 for a quicker first token, open weights you can run yourself and more hosts to choose from on OpenRouter (30 against 3). Choose GPT-6 Luna for the lower cost per correct answer ($0.00034 against $0.0012), faster streaming and shorter outputs (about 431 tokens per question against 1,318). The widest subject gap is in medicine: GPT-6 Luna answered 6 of 11 correctly, DeepSeek V4.1 Flash 1017 4 of 11. With so few questions per subject, treat this as a lead to test on your own prompts, not a result.
Accuracy gap +1.3 points (z = 0.18). We call a winner only when z is at least 1.96, the usual 95% threshold. Speed, latency and price need a 10% margin.