Verdict
Built from the measurements below
- Gemini 3.5 Flash Lite is more accurate: 75.0% against 30.0% on the same 80 questions, a 45.0-point gap that is statistically significant.
- Gemma 4 31B is 2.4× cheaper per question as billed ($0.0014 vs $0.0034), reasoning tokens included.
- Gemini 3.5 Flash Lite generates text 7.0× faster: 409 tokens per second against 58.
- Gemma 4 31B starts answering sooner: 0.6 s to the first token, against 1.1 s for Gemini 3.5 Flash Lite.
- Gemini 3.5 Flash Lite accepts more context: 1M tokens against 262k.
Which should you choose?
Choose Gemini 3.5 Flash Lite for the most accurate answers of the two, faster streaming, shorter outputs (about 1,311 tokens per question against 3,307) and a 1M-token context window. Choose Gemma 4 31B for a quicker first token, open weights you can run yourself and more hosts to choose from on OpenRouter (13 against 2). The widest subject gap is in engineering: Gemini 3.5 Flash Lite answered 10 of 15 correctly, Gemma 4 31B 0 of 15. With so few questions per subject, treat this as a lead to test on your own prompts, not a result.
Accuracy gap +45.0 points (z = 5.70). We call a winner only when z is at least 1.96, the usual 95% threshold. Speed, latency and price need a 10% margin.