Verdict
Built from the measurements below
- Mistral Large 4 is more accurate: 50.0% against 26.2% on the same 80 questions, a 23.8-point gap that is statistically significant.
- Ministral 3 3B 2512 is 58× cheaper per question as billed ($0.00019 vs $0.011), reasoning tokens included.
- Ministral 3 3B 2512 generates text 3.0× faster: 252 tokens per second against 83.
- Ministral 3 3B 2512 starts answering sooner: 0.3 s to the first token, against 1.1 s for Mistral Large 4.
- Mistral Large 4 accepts more context: 1M tokens against 131k.
Which should you choose?
Choose Ministral 3 3B 2512 for the lower cost per correct answer ($0.00074 against $0.023), faster streaming, a quicker first token, shorter outputs (about 1,652 tokens per question against 2,613) and open weights you can run yourself. Choose Mistral Large 4 for the most accurate answers of the two and a 1M-token context window. The widest subject gap is in science: Mistral Large 4 answered 20 of 30 correctly, Ministral 3 3B 2512 5 of 30. With so few questions per subject, treat this as a lead to test on your own prompts, not a result.
Accuracy gap -23.8 points (z = -3.09). We call a winner only when z is at least 1.96, the usual 95% threshold. Speed, latency and price need a 10% margin.