Verdict
Built from the measurements below
- Accuracy is too close to call: Kimi K2.7 Code scored 65.0% and Kimi K3.1 Pro 77.5% on the same 80 questions, a gap within what chance alone could produce.
- Kimi K2.7 Code is 1.7× cheaper per question as billed ($0.006 vs $0.01), reasoning tokens included.
- Kimi K2.7 Code generates text 3.6× faster: 136 tokens per second against 38.
- Kimi K2.7 Code starts answering sooner: 0.4 s to the first token, against 1.3 s for Kimi K3.1 Pro.
- Kimi K3.1 Pro accepts more context: 1M tokens against 262k.
Which should you choose?
Choose Kimi K2.7 Code for the lower cost per correct answer ($0.0092 against $0.014), faster streaming, a quicker first token, open weights you can run yourself and more hosts to choose from on OpenRouter (11 against 1). Choose Kimi K3.1 Pro for shorter outputs (about 973 tokens per question against 1,781) and a 1M-token context window. The widest subject gap is in engineering: Kimi K3.1 Pro answered 11 of 15 correctly, Kimi K2.7 Code 7 of 15. With so few questions per subject, treat this as a lead to test on your own prompts, not a result.
Accuracy gap -12.5 points (z = -1.75). We call a winner only when z is at least 1.96, the usual 95% threshold. Speed, latency and price need a 10% margin.