Verdict
Built from the measurements below
- Accuracy is too close to call: Command A scored 45.0% and Mistral Large 4 50.0% on the same 80 questions, a gap within what chance alone could produce.
- Command A is 1.6× cheaper per question as billed ($0.007 vs $0.011), reasoning tokens included.
- Mistral Large 4 generates text 12× faster: 83 tokens per second against 7.
- Mistral Large 4 starts answering sooner: 1.1 s to the first token, against 1.8 s for Command A.
- Mistral Large 4 accepts more context: 1M tokens against 256k.
Which should you choose?
Choose Command A for the lower cost per correct answer ($0.015 against $0.023), shorter outputs (about 616 tokens per question against 2,613) and open weights you can run yourself. Choose Mistral Large 4 for faster streaming, a quicker first token and a 1M-token context window. The widest subject gap is in medicine: Command A answered 6 of 11 correctly, Mistral Large 4 2 of 11. With so few questions per subject, treat this as a lead to test on your own prompts, not a result.
Accuracy gap -5.0 points (z = -0.63). We call a winner only when z is at least 1.96, the usual 95% threshold. Speed, latency and price need a 10% margin.