Verdict
Built from the measurements below
- gpt-oss-20b is more accurate: 51.2% against 25.0% on the same 80 questions, a 26.2-point gap that is statistically significant.
- Granite 4.0 Micro is 1.3× cheaper per question as billed ($0.000051 vs $0.000066), reasoning tokens included.
- gpt-oss-20b generates text 2.7× faster: 103 tokens per second against 38.
Which should you choose?
Choose gpt-oss-20b for the most accurate answers of the two, the lower cost per correct answer ($0.00013 against $0.0002), faster streaming and more hosts to choose from on OpenRouter (10 against 1). The widest subject gap is in economics: gpt-oss-20b answered 6 of 9 correctly, Granite 4.0 Micro 2 of 9. With so few questions per subject, treat this as a lead to test on your own prompts, not a result.
Accuracy gap +26.2 points (z = 3.42). We call a winner only when z is at least 1.96, the usual 95% threshold. Speed, latency and price need a 10% margin.